Pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method

Abstract

The invention relates to the technical field of medical treatment, in particular to a pre-hospital and in-hospital first-aid time axis dynamic tracking and analysis method, which comprises the following steps of: acquiring patient vital sign data and a chief complaint voice stream transmitted by a pre-hospital first-aid terminal in real time, inputting a special voice recognition engine, and converting to generate structured text data; extracting a patient identity label and an illness state characteristic field and writing into a time axis database; and in-hospital medical equipment operation interface signals, geo-fence trigger events and triage system events are captured in real time, and diagnosis and treatment node data with timestamps are generated and added to the time axis database. According to the method, historical node sequences in a time axis database are extracted to construct a disease feature vector set, incremental training samples are obtained after feature sorting, the progress deviation value of real-time node data is combined to match historical similar case time axis fragments to execute weighted deduction, and a prediction time sequence of a key link is generated. And a reliable basis can be provided for medical personnel to plan resource scheduling of equipment, personnel and the like in advance.

Description

Technical Field

[0001] This invention relates to the field of medical technology, and in particular to a method for dynamic tracking and analysis of the timeline of pre-hospital and in-hospital emergency care. Background Technology

[0002] Pre-hospital and in-hospital emergency care is a continuous process encompassing multiple stages, including on-site patient assistance, transfer, and in-hospital reception, examination, and treatment. It involves multi-source data such as vital signs, chief complaints, equipment operating status, and system events. Accurate recording and correlation of time nodes at each stage are of great value in maintaining the continuity of the emergency care process and ensuring the orderly progress of treatment.

[0003] The timeline serves as a core carrier connecting multiple stages before and during hospitalization and integrating multi-source data. It accurately records time nodes for each stage, such as the start time of on-site rescue, the departure and arrival time of transfer, and the start time of in-hospital admission and examination, and establishes logical connections between nodes to form a complete emergency rescue process time chain.

[0004] Existing technologies lack time prediction capabilities for key treatment steps in pre-hospital and in-hospital emergency care processes. It is difficult to predict the timing of critical nodes such as the first medical contact and the initiation of examinations in advance, making it difficult for medical staff to plan resource allocation in advance, which can easily lead to poor coordination or delays in treatment. Summary of the Invention

[0005] To overcome the above shortcomings, this invention provides a method for dynamic tracking and analysis of the timeline of pre-hospital and in-hospital emergency care. It aims to improve the existing technology's lack of time prediction for key treatment links, making it difficult to predict the time of key nodes such as the first medical contact and the initiation of examinations in advance, which can easily lead to poor connection or delays in treatment links.

[0006] In a first aspect, the present invention provides the following technical solution: a method for dynamic tracking and analysis of the timeline of pre-hospital and in-hospital emergency care, comprising the following steps: The system acquires patient vital signs data and chief complaint voice stream transmitted from pre-hospital emergency terminals in real time, inputs them into a dedicated speech recognition engine to generate structured text data, and extracts patient identification and disease characteristic fields to write them into a timeline database. Real-time capture of medical equipment operation interface signals, geofence trigger events, and triage system events within the hospital; generate timestamped diagnosis and treatment node data and append it to the timeline database. The system calls upon historical case data from the timeline database, integrates real-time hospital resource status data to train a prediction model, and outputs the predicted time series of key treatment steps to the dynamic threshold engine. The pre-stored standard time consumption benchmark value for disease is retrieved and input into the dynamic threshold engine, and compared with the predicted time series. When the deviation exceeds the limit, a graded early warning instruction is triggered. A structured early warning message containing the location identifier of the delay is generated and pushed to the medical staff terminal. The intervention execution data received from the terminal is written back to the timeline database.

[0007] Furthermore, the real-time acquisition includes: Noise reduction processing is performed on the voice stream inside the ambulance cabin, and the medical entity recognition model is input to output the patient's condition feature data; Analyze patient identification document images to obtain identification data; The patient's structured record is generated by integrating the disease characteristic data and identity identification data and written into the patient information table; When the voice command to start the transfer is recognized, the patient's structured record is extracted, the current timestamp is bound to generate an initial timeline node, and the timeline database is written.

[0008] Furthermore, the real-time capture of events within the hospital includes: The system receives GPS coordinates from the ambulance as input to the geofence mapping engine and outputs a fence trigger signal to generate an arrival event record. Parse the medical device operation code to obtain the DICOM operation status, convert it into a standard operation event, and associate it with the current patient identifier; Subscribe to status change messages in the laboratory information system and capture report review completion notifications to generate test readiness events; The hospital arrival event records, standard operation events, and test readiness events are integrated, and timestamps are bound to generate diagnosis and treatment node data, which is then written to the timeline database.

[0009] Furthermore, the training prediction model includes: extracting historical node sequences from the timeline database to construct a disease feature vector set and inputting it into the feature sorter; Perform feature importance ranking and output a list of core features to the model trainer; Based on the core feature list, an incremental training sample set is obtained by applying a dynamic time window sliding update mechanism. Incremental training is performed to generate optimized model parameters, which are then output to the prediction engine.

[0010] Furthermore, the dynamic threshold engine operation includes: Real-time data collection of the number of patients waiting in the emergency department and the ratio of attending physicians on duty; calculation of resource shortage index. Based on the resource shortage index, the pre-stored standard time benchmark values ​​for disease types are retrieved, and a dynamic floating algorithm is applied to generate a graded early warning threshold. The graded early warning threshold is output to the early warning comparison device for real-time monitoring and comparison.

[0011] Furthermore, the generation of the predicted time series includes: Real-time scanning of the latest node data in the timeline database; input deviation calculator outputs progress deviation. Based on the aforementioned progress deviation, a weighted time extrapolation is performed by matching timeline segments of similar historical cases. The system generates predicted values ​​for the first medical contact time, examination initiation time, and treatment start time, forming a predicted time series which is then output to the early warning system.

[0012] Furthermore, the triggering of tiered early warnings includes: Receive signals exceeding limits, classify warning levels, and generate warning level identifiers; Extract related node data from the timeline database to construct a topology diagram of delay links; Perform root cause analysis by linking the historical root cause database and output influencing factor analysis data; The system integrates the warning level identifier, the delay path topology diagram, and the influencing factor analysis data to generate a visualized warning report, which is then pushed to the medical staff terminal.

[0013] Furthermore, the root cause analysis includes: Real-time retrieval of equipment occupancy rate and contribution model of human resource data input; Construct a multi-factor Bayesian network contribution inference model and output a ranked list; Write the sorting list into the warning report.

[0014] Furthermore, the intervention execution data fed back by the receiving terminal includes: Resource scheduling instructions are obtained through the terminal interface and input into the execution recorder; Write the instruction execution timestamp, operator identification, and resource change status into the intervention field to associate the delay link location identifier; Calculate the deviation between the actual and predicted time. When the deviation exceeds a set threshold, it is marked as an optimization case and used as an incremental training sample to input into the prediction model training process.

[0015] Secondly, the present invention provides the following technical solution: a pre-hospital and in-hospital emergency care timeline dynamic tracking and analysis system, comprising: The multi-source data acquisition module is used to acquire pre-hospital and in-hospital data and convert them into structured diagnosis and treatment node data; The timeline database is connected to the multi-source data acquisition module, stores data, and constructs a patient-specific emergency timeline. The predictive analysis module connects to the timeline database, trains the model, and outputs the predicted time of key processes. The dynamic threshold engine, connected to the prediction and analysis module, compares the predicted time with the standard time, and triggers an early warning if the deviation exceeds the limit. The early warning interaction module connects the dynamic threshold engine and the timeline database, pushes early warnings, and receives intervention feedback to write back to the database.

[0016] The present invention has the following beneficial effects: 1. In this invention, a disease feature vector set is constructed by extracting historical node sequences from the timeline database. After feature sorting, incremental training samples are obtained based on a dynamic time window sliding update mechanism. Incremental training is used to optimize model parameters. Combined with the progress deviation of real-time node data, weighted extrapolation is performed by matching timeline segments of similar historical cases to generate predicted time series of key links. This can provide a reliable basis for medical staff to plan the scheduling of equipment, personnel and other resources in advance, reduce resource mismatch or delays, and improve emergency rescue efficiency.

[0017] 2. In this invention, pre-hospital patient vital signs and chief complaints are acquired in real time and converted into structured data through noise reduction, speech recognition, and medical entity recognition. Initial timeline nodes are generated by combining patient identification. At the same time, events such as ambulance arrival, medical equipment operation status, and test report readiness are captured in real time to generate time-stamped diagnosis and treatment nodes. These are uniformly written into the timeline database to construct a patient-specific emergency timeline, ensuring the continuous recording and accurate correlation of emergency data from pre-hospital to in-hospital, and providing complete and reliable data support for full-process time management.

[0018] 3. In this invention, the dynamic threshold engine calculates the resource shortage index based on the ratio of emergency patients to attending physicians on duty, and generates a dynamic graded early warning threshold by combining the standard time benchmark value of the disease. When the threshold is exceeded, a graded early warning is triggered and a visual report containing the location of delay links and root cause analysis is pushed. At the same time, the intervention execution data of medical staff terminals is received and written back to the database. Cases that exceed the deviation are marked as optimization samples to feed back the prediction model training, forming a continuous iterative closed loop from early warning to intervention and then to model optimization. This is conducive to timely detection of delay risks and implementation of intervention, and ensures the orderly progress of the emergency process. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for dynamic tracking and analysis of the timeline of pre-hospital and in-hospital emergency care proposed in this invention; Figure 2 This is an architecture diagram of a pre-hospital and in-hospital emergency timeline dynamic tracking and analysis system proposed in this invention. Detailed Implementation

[0020] The technical solutions in 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.

[0021] Example 1 In a first embodiment of the present invention, the present invention provides a method for dynamic tracking and analysis of the timeline of pre-hospital and in-hospital emergency care, such as... Figure 1 As shown, it includes the following steps: The system acquires patient vital signs data and chief complaint voice stream transmitted from pre-hospital emergency terminals in real time, inputs them into a dedicated speech recognition engine to generate structured text data, and extracts patient identification and disease characteristic fields to write them into a timeline database. Furthermore, real-time acquisition includes: Noise reduction processing is performed on the voice stream inside the ambulance cabin, and the medical entity recognition model is input to output the patient's condition feature data; Analyze patient identification document images to obtain identification data; The patient's structured record is generated by integrating disease characteristic data and identity identification data and written into the patient information table; When the voice command to start the transfer is recognized, the patient's structured record is extracted, the current timestamp is bound to generate an initial timeline node, and the timeline database is written to it.

[0022] Specifically, a multi-channel microphone array deployed inside the ambulance cabin collects the spoken language of medical staff and patients, while connected vital sign monitoring equipment simultaneously transmits data such as heart rate, blood pressure, and blood oxygen saturation. The spoken language is first processed by an adaptive noise reduction algorithm based on minimum mean square error. This algorithm uses noisy spoken language signals... As input, the weight vector is updated iteratively. Make output With the expectation of clean speech error The mean square value is the smallest, that is in and To remove environmental noise interference, the noise-reduced speech stream is input into a dedicated speech recognition engine based on deep learning and converted into text. It is then input into a BERT named entity recognition model trained on medical corpus. This model takes the text sequence as input and outputs disease feature data labeled with symptoms such as chest pain and shortness of breath, and signs such as blood pressure of 150 / 90 mmHg. Simultaneously, the image of the patient's ID card is captured by a high-definition camera, and the name, ID number and other identity identification data are parsed using OCR recognition technology based on convolutional neural networks. Through field matching, the disease feature data and identity identification data are fused to generate a structured record containing the patient's basic information and initial condition, which is written into the patient information table. When the dedicated speech recognition engine recognizes the preset voice command to start the transfer, it automatically extracts the patient's structured record, binds it to the current time to generate an initial timeline node containing the patient's identity, initial condition and transfer start time, and writes it into the timeline database. This enables real-time acquisition and structured processing of pre-hospital emergency data, ensuring accurate association and timely recording of patient identity and medical information, and providing initial data support for the construction and tracking of subsequent emergency timelines.

[0023] Real-time capture of signals from the operation interfaces of medical equipment within the hospital, geofence-triggered events, and triage system events; generate timestamped treatment node data and append it to the timeline database. Furthermore, real-time capture of events within the facility includes: The system receives GPS coordinates from the ambulance as input to the geofence mapping engine and outputs a fence trigger signal to generate an arrival event record. Parse the medical device operation code to obtain the DICOM operation status, convert it into a standard operation event, and associate it with the current patient identifier; Subscribe to status change messages in the laboratory information system and capture report review completion notifications to generate test readiness events; The system integrates hospital event records, standard operating procedure events, and laboratory readiness events, binds timestamps to generate diagnosis and treatment node data, and writes it into the timeline database.

[0024] Specifically, the hospital receives the latitude and longitude coordinates output by the ambulance's GPS in real time via a dedicated interface, inputs them into a geofencing engine, and the engine uses a ray-mapping method to determine whether the coordinates are within the hospital's preset geographical area. Specifically, the ambulance's real-time coordinates are set as points... The hospital's geofence is centered around the following vertices: , , , A polygon, from point Draw a horizontal ray to the right, and calculate the intersection of this ray with each side of the polygon. ( The number of intersections of ) ,like If the number is odd, then the decision point is... Inside the fence, a fence trigger signal is output and an arrival event record is generated. Simultaneously, the system connects to the DICOM protocol interface of the medical equipment, parses the status fields such as examination start flag and image acquisition completion flag in the equipment operation code in real time, and maps them to standardized examination operation events such as CT scan start and MRI completion. The system also associates the current emergency patient through the patient ID field. In addition, it connects to the laboratory information system through the HL7 message subscription mechanism. When a notification of the completion status change of the test report is received, the patient ID and review time corresponding to the report are extracted to generate a test ready event. Finally, the arrival event record, standard operation event, and test ready event are aligned, and the system timestamp of each event is uniformly bound. The data is integrated into diagnosis and treatment node data containing event type, associated patient ID, and occurrence time, and appended to the timeline database. This enables real-time capture and standardized recording of critical medical events within the hospital, ensuring the continuity and integrity of the emergency timeline and providing consistent node data support for subsequent time tracking and analysis.

[0025] Historical case data from the timeline database is called up and real-time hospital resource status data is integrated to train a prediction model, and the predicted time series of key treatment links are output to the dynamic threshold engine. Furthermore, training the prediction model includes: extracting historical node sequences from the timeline database to construct a disease feature vector set, which is then input into the feature sorter; Perform feature importance ranking and output a list of core features to the model trainer; Based on the core feature list, an incremental training sample set is obtained by applying a dynamic time window sliding update mechanism. Incremental training is performed to generate optimized model parameters, which are then output to the prediction engine.

[0026] Furthermore, generating the predicted time series includes: Real-time scanning of the latest node data in the timeline database; input deviation calculator outputs progress deviation. Based on the progress deviation, perform weighted time extrapolation by matching timeline segments of similar historical cases; The system generates predicted values ​​for the first medical contact time, examination initiation time, and treatment start time, forming a predicted time series which is then output to the early warning system.

[0027] Specifically, historical node sequences corresponding to each disease are extracted from the timeline database, and features such as patient gender, disease type, pre-hospital transfer time, and vital signs upon arrival are extracted to construct a disease feature vector set, which is then input into a feature sorter. This sorter uses a random forest algorithm to calculate feature importance, quantifying the degree of influence of node features on the model's prediction error. Its importance For all decision trees, due to the use The average decrease in the Gini index caused by splitting. ,in For the number of trees, For the first Among the trees, due to characteristics The reduction in the Gini index caused by the split is used to output a list of core features, including pre-hospital transport time and the number of current emergency room doctors on duty, to the model trainer. Based on this list of core features, a dynamic time window sliding update mechanism is applied. The window start time is automatically adjusted as the current emergency event progresses, and the window contains historical case data similar to the current case. Newly generated diagnosis and treatment node data is continuously incorporated through the sliding window to obtain an incremental training sample set. Incremental training is performed using gradient descent. Model parameters... The update formula is ,in For learning rate, For the first Incremental samples The corresponding loss function gradient is used to generate optimized model parameters, which are then output to the prediction engine. During the prediction phase, the latest node data, such as the arrival time of currently completed hospital events, is scanned in real-time from the timeline database. This data is input into the deviation calculator to calculate the difference between the actual occurrence time of this node and the average time of the same disease and stage in history. This difference is used as the progress deviation. Based on this deviation, timeline segments of similar historical cases are matched using cosine similarity. ,in For the current case feature vector, Using historical case feature vectors, the matched segments are weighted according to similarity to perform weighted time extrapolation, and finally the predicted values ​​of the first medical contact time, examination start time, and treatment start time are generated to form a predicted time series and output to the early warning system. By dynamically incorporating new data to continuously optimize model parameters, the adaptability to emergency procedures for different diseases is improved. By matching similar cases, the scenario fit of time prediction is enhanced, and the deviation between the predicted time and the actual time of key links such as the first medical contact is controlled within a narrower range. This provides a reliable basis for medical staff to plan resource allocation in advance. At the same time, the data on the deviation between predicted and actual time accumulated in the timeline database feeds back into the feature ranking and model training process, forming a continuous iterative optimization loop. This continuously improves the model's adaptability in complex emergency scenarios and effectively supports time management and efficiency improvement in emergency procedures.

[0028] The pre-stored standard time baseline value for disease types is retrieved and input into the dynamic threshold engine, compared with the predicted time series, and a graded early warning instruction is triggered when the deviation exceeds the limit. Furthermore, the dynamic threshold engine operation includes: Real-time data collection of the number of patients waiting in the emergency department and the ratio of attending physicians on duty; calculation of resource shortage index. Based on the resource scarcity index, the pre-stored standard time benchmark values ​​for disease types are retrieved, and a dynamic floating algorithm is applied to generate graded early warning thresholds. Output the graded early warning threshold to the early warning comparison device for real-time monitoring and comparison.

[0029] Furthermore, triggering tiered early warnings includes: Receive signals exceeding limits, classify warning levels, and generate warning level identifiers; Extract related node data from the timeline database to construct a topology diagram of delay links; Perform root cause analysis by linking the historical root cause database and output influencing factor analysis data; The system integrates early warning level indicators, delay path topology diagrams, and influencing factor analysis data to generate a visualized early warning report, which is then pushed to medical staff terminals.

[0030] Furthermore, root cause analysis includes: Real-time retrieval of equipment occupancy rate and contribution model of human resource data input; Construct a multi-factor Bayesian network contribution inference model and output a ranked list; Write it into the sorting list and add it to the warning report.

[0031] Specifically, the dynamic threshold engine continuously collects data on the number of patients waiting in the emergency room and the number of attending physicians on duty through the hospital information system's real-time interface. The data synchronization frequency is dynamically adjusted according to the busyness of the emergency room, and the synchronization interval is shortened during peak hours to ensure timeliness. Based on this, the resource shortage index is calculated. ,in ; This represents the current number of patients remaining in the emergency room. The number of attending physicians on duty. The historical average allocation ratio for this period was used as the basis for calculation. The results were normalized and mapped to the 0-1 range, with higher values ​​indicating greater resource scarcity. Based on this index, the baseline value for the standard time consumption of each disease was retrieved. A dynamic floating algorithm is applied to generate tiered early warning thresholds, with stricter floating coefficients used for critical illnesses, such as when... Time threshold is ,when Time threshold is ,when Time threshold is To ensure adequate buffering during periods of extreme resource strain without ignoring critical delays, the output tiered warning thresholds are pushed to the warning comparison device in real time. The comparison device continuously compares the predicted time series with the thresholds. Once an out-of-limit signal is detected, a warning level is assigned based on the deviation magnitude, and a corresponding identifier is generated. Mild deviations generate a Level 1 identifier, and significant deviations generate a Level 2 identifier. Different levels of identifiers are associated with different terminal push priorities. Simultaneously, data on diagnosis and treatment nodes related to the current patient are extracted from the timeline database. A topology map of delay links is constructed based on chronological order and node associations, visually presenting the connection relationships between each link and the path of accumulated delays. The root cause analysis stage retrieves real-time occupancy rates of equipment such as CT and ultrasound, as well as human resource allocation data such as emergency nurses and technicians. This data is input into a pre-trained multi-factor Bayesian network contribution inference model. This model is trained based on the correlation data between various factors and delay outcomes in historical delay events, calculating the posterior probability of each factor. ( Factors such as excessive equipment occupancy and insufficient nurse ratio have contributed to this situation. For the current delay event, determine the contribution of each factor to the delay and output a list of influencing factors sorted by contribution; finally, integrate the warning level label, the delay link topology map and the list of influencing factors to generate a warning report that includes time axis visualization and root cause percentage chart, and push it through multiple channels such as the medical and nursing terminal message center and mobile nursing APP according to the warning level. By dynamically adapting the resource scarcity index to the early warning threshold, the early warning standards are accurately matched with the real-time medical resource status, avoiding misjudgments or omissions caused by resource fluctuations. Bayesian network root cause analysis transforms scattered equipment and personnel data into quantifiable ranking of influencing factors. Combined with the intuitive presentation of the topology map, it helps medical staff quickly locate the source of delays and transmission paths. Different levels of early warning push mechanisms ensure that key information reaches the corresponding level of personnel first, effectively supporting the timely detection and targeted intervention of delay risks in the emergency process, and ensuring the orderly progress of the emergency response.

[0032] The system generates structured early warning messages containing location identifiers of delays and pushes them to medical staff terminals. It also receives intervention execution data from the terminals and writes it back to the timeline database. Furthermore, the intervention execution data received from the receiving terminal includes: Resource scheduling instructions are obtained through the terminal interface and input into the execution recorder; Write the instruction execution timestamp, operator identification, and resource change status into the intervention field to associate the delay link location identifier; Calculate the deviation between the actual and predicted time. When the deviation exceeds a set threshold, it is marked as an optimization case and used as an incremental training sample to input into the prediction model training process. Specifically, structured early warning messages integrate delayed link location identifiers, early warning levels, and key time node information in a standardized data format. These messages are pushed through the real-time communication interface of medical staff terminals. The terminal interface prominently displays the specific location of the delayed link on the timeline. The terminal's interactive interface listens for resource scheduling commands triggered by medical staff, such as equipment allocation requests and personnel reinforcement requests. The command content is input into the execution recorder for structured parsing. The recorder automatically captures the system timestamp and operator ID at the time of command execution, and simultaneously collects resource status change data, such as CT equipment changing from occupied to ready, and nurses changing from busy to idle. This information is written into the intervention field of the timeline database and associated with the corresponding early warning event through the delayed link location identifier. Subsequently, the deviation between the actual completion time and the predicted time of that link is calculated. ,when When the deviation threshold set based on historical intervention effects is exceeded, it is automatically marked as an optimization case. The intervention measures, resource change curves and time deviation data of the case are extracted and incorporated into the training dataset of the prediction model as incremental training samples to participate in the model parameter update. This achieves closed-loop data recording of the entire early warning and intervention process, making each intervention operation traceable and accurately linked to the delay links. The optimized incremental training mechanism of cases allows the prediction model to continuously absorb actual intervention experience, gradually improving the fit of time node prediction in complex emergency scenarios, and providing more adaptive decision support for time management in subsequent emergency procedures.

[0033] Example 2: In the pre-hospital and in-hospital emergency care process, the lack of a time prediction mechanism for key treatment steps makes it difficult to anticipate the timing of critical nodes such as the initial medical contact and the initiation of examinations. This makes it difficult for medical staff to plan resource allocation in advance, easily leading to poor coordination or delays in treatment processes. For example Figure 2 As shown, to solve the above problems, the present invention provides a pre-hospital and in-hospital emergency care timeline dynamic tracking and analysis system, comprising: The multi-source data acquisition module is used to acquire pre-hospital and in-hospital data and convert them into structured diagnosis and treatment node data; The timeline database connects to multi-source data acquisition modules, stores data, and constructs a patient-specific emergency timeline. The predictive analysis module connects to a timeline database, trains a model, and outputs predicted times for key processes. The dynamic threshold engine connects to the predictive analysis module, compares the predicted time with the standard time, and triggers an alert when the deviation exceeds the limit. The early warning interaction module connects the dynamic threshold engine and the timeline database, pushes early warnings, and receives intervention feedback to write back to the database.

[0034] Specifically, the multi-source data acquisition module acquires the patient's vital signs and chief complaint voice stream through the pre-hospital emergency terminal interface, and converts them into structured data through noise reduction, speech recognition and medical entity recognition. Simultaneously, it collects the GPS coordinates of the hospital ambulance, the DICOM operation signals of medical equipment and the status of the laboratory information system, and converts them into diagnosis and treatment node data with timestamps. The timeline database connects to the multi-source data acquisition module, integrating pre-hospital initial records and various in-hospital diagnosis and treatment nodes according to patient identification, and constructing a dedicated emergency timeline that includes events such as transfer initiation, arrival at the hospital, and examination procedures; The predictive analysis module retrieves historical cases and real-time resource data from the timeline database, generates a predictive model through feature sorting and dynamic time window incremental training, and outputs the predicted time of key steps such as the first medical contact and the start of the examination. The dynamic threshold engine receives the predicted time, combines the standard time baseline value for the disease with the resource shortage index to generate a dynamic early warning threshold, and triggers an early warning for exceeding the limit after comparison. The early warning interaction module pushes structured early warnings containing delay location to medical staff terminals, receives intervention instructions such as resource scheduling, records execution timestamps, operator and resource status changes and writes them back to the timeline database, and marks cases that exceed the deviation as optimized samples to feed back into the prediction model training.

[0035] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method, characterized in that, The method comprises the following steps: Real-time acquisition of patient vital sign data and chief complaint voice stream transmitted by pre-hospital emergency terminal, input of special voice recognition engine to convert and generate structured text data, extraction of patient identity and disease characteristic field and writing into time axis database; Real-time capture of in-hospital medical device operation interface signal, geographic fence triggering event and triage system event, generation of time-stamped diagnosis and treatment node data and appending to the time axis database; Calling historical case data in the time axis database, fusing real-time hospital resource state data to train prediction model, and outputting prediction time sequence of key treatment link to dynamic threshold engine; Calling pre-stored disease standard time consumption benchmark value input into the dynamic threshold engine, comparing with the prediction time sequence, and triggering graded early warning instruction when deviation exceeds limit; Generating structured early warning message containing delay link positioning identifier and pushing to medical terminal, receiving terminal feedback intervention execution data and writing back to the time axis database.

2. The pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method according to claim 1, characterized in that, The real-time acquisition comprises: Noise reduction processing on voice stream in ambulance, input of medical entity recognition model to output disease characteristic data; Analyzing patient identity card image to obtain identity data; Fusing the disease characteristic data and identity data to generate patient structured record and writing into patient information table; When a transfer start voice instruction is recognized, the patient structured record is extracted, the current timestamp is bound to generate an initial time axis node and written into the time axis database.

3. The pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method according to claim 1, characterized in that, The real-time capture of in-hospital events comprises: Receiving ambulance global positioning system coordinates input into geographic fence mapping engine, outputting fence triggering signal to generate in-hospital event record; Analyzing medical device operation code to obtain DICOM operation state and converting it into standard operation event associated with current patient identifier; Subscribing to laboratory information system state change message, capturing report review completion notification to generate test ready event; Fusing the in-hospital event record, standard operation event and test ready event, binding timestamp to generate diagnosis and treatment node data and writing into time axis database.

4. The pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method according to claim 1, characterized in that, The training prediction model comprises: Extracting historical node sequence in time axis database to construct disease characteristic vector set and inputting into feature sorter; Performing feature importance sorting to output core feature list to model trainer; Based on the core feature list, applying dynamic time window sliding update mechanism to obtain incremental training sample set; 5. The pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method according to claim 1, characterized in that, Performing incremental training to generate optimized model parameters and outputting to prediction engine. The dynamic threshold engine operation comprises: Real-time collection of emergency retention patient quantity and on-duty attending physician ratio data, calculation of resource shortage index; Based on the resource shortage index, calling pre-stored disease standard time consumption benchmark value and applying dynamic floating algorithm to generate graded early warning threshold; 6. The pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method according to claim 1, characterized in that, Outputting the graded early warning threshold to early warning comparator to perform real-time monitoring comparison. The prediction time sequence generation comprises: Real-time scanning of latest node data in time axis database and inputting into deviation calculator to output progress deviation amount; Based on the progress deviation amount, matching time axis segments of similar historical cases to perform weighted time deduction; 7. The pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method according to claim 1, characterized in that, Generating first medical contact time prediction value, examination start time prediction value and treatment start time prediction value to form prediction time sequence and outputting to early warning system. The graded early warning triggering comprises: Receiving an over-limit signal to divide an early warning level to generate an early warning level identifier; Extracting associated node data in the time axis database to construct a delay link topology graph; Associated historical root cause library to perform root cause analysis and output impact factor analysis data; Fusion of the early warning level identifier, delay link topology graph, and impact factor analysis data to generate a visual early warning report and push it to a medical terminal.

8. The pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method according to claim 7, characterized in that, The root cause analysis includes: Real-time retrieval of device occupancy rate and human resource data input contribution model; Constructing a multi-factor Bayesian network contribution reasoning model to output a ranking list; Write the ranking list to the early warning report.

9. The pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method according to claim 1, characterized in that, The intervention execution data received from the terminal feedback includes: Obtain resource scheduling instructions through the terminal interface and input execution recorder; Write instruction execution timestamp, operator identity, and resource change state to the intervention field associated with the delay link positioning identifier; Calculate the actual and predicted time deviation, and when the deviation exceeds the set threshold, mark it as an optimization case as an incremental training sample input into the prediction model training process.

10. A pre-hospital and in-hospital emergency time axis dynamic tracking and analysis system, characterized in that, A pre-hospital and in-hospital emergency time axis dynamic tracking and analysis method according to any one of claims 1-9, comprising: A multi-source data acquisition module for acquiring pre-hospital and in-hospital data and converting it into structured diagnosis and treatment node data; A time axis database connected to the multi-source data acquisition module, storing data and constructing a patient-specific emergency time axis; A prediction analysis module connected to the time axis database, training the model to output the predicted time of the key link; A dynamic threshold engine connected to the prediction analysis module, comparing the predicted time with the standard time consumption, and triggering an early warning when the deviation exceeds the limit; An early warning interaction module connected to the dynamic threshold engine and the time axis database, pushing the early warning and receiving intervention feedback to rewrite the database.

Images