Nursing shift change system based on homologous heterogeneous data fusion and large language model
The nursing handover system, which integrates homogeneous and heterogeneous data and a large language model, solves the problems of information omission and time consumption in nursing handover, realizes unified data expression and identification of key events, and generates clear nursing handover reports.
Patent Information
- Application Number
- CN202610238824.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-28
- Publication Date
- 2026-03-27
- Estimated Expiration
- 2046-02-28
AI Technical Summary
Existing nursing handover systems suffer from problems such as easy omission of key handover information, high error rate, long processing time, scattered data sources with inconsistent structures, and lack of knowledge-level integration, semantic alignment, and closed-loop quality control.
A nursing handover system based on homogeneous heterogeneous data fusion and a large language model is adopted. The data fusion module performs data alignment and semantic alignment, the event sequence modeling module constructs the logical association of clinical events, the key event screening module calculates scores and generates nursing handover reports, and the large language model generates global summaries and detailed handover information.
It enables a unified representation of structured and unstructured data, reduces information fragmentation, identifies key events, generates clear shift handover reports, and improves handover efficiency and information accuracy.
Smart Images

Figure CN121743482A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of software technology, and in particular to a nursing shift handover system based on the fusion of homogeneous and heterogeneous data and a large language model. Background Technology
[0002] In current nursing handover procedures, nurses need to manually retrieve, verify, and transcribe information across multiple systems such as HIS / EMR / LIS / PACS / CIS / IoT, resulting in: 1. Key information during shift handover is easily overlooked, and the error rate is high; 2. Shift handover is time-consuming, and indirect nursing time accounts for a large proportion; 3. The data sources are scattered and the structures are inconsistent (homogeneous but heterogeneous), making it difficult to form a unified panoramic view of patients.
[0003] Existing systems mostly focus on displaying data, lacking knowledge-level integration, semantic alignment, risk focus, and closed-loop quality control. Summary of the Invention
[0004] The main objective of this invention is to address the technical problems of existing technologies lacking knowledge-level fusion, semantic alignment, risk focusing, and closed-loop quality control. A nursing shift handover system based on homogeneous heterogeneous data fusion and a large language model includes: The data fusion module is used to obtain structured and unstructured data of the same patient from multiple clinical business systems, perform unified data model alignment and semantic alignment on the structured and unstructured data, and construct a panoramic data view of the patient based on a unified timeline. The event sequence modeling module is used to segment the patient panoramic data view into multiple clinical events based on the data changes at adjacent time points in the patient panoramic data view, and to construct the logical association strength of the clinical events based on the temporal correlation and semantic similarity between the clinical events. The critical event screening module is used to calculate the anomaly score of numerical data and the semantic fluctuation score of text data for the clinical events, respectively, and to perform a weighted fusion of the anomaly score and the semantic fluctuation score to screen out critical events from the clinical events according to the fusion score. The shift handover generation module includes a first generation unit and a second generation unit, wherein: The first generation unit is used to generate a global summary of nursing handover information by calling a large language model based on all the clinical events. The second generation unit is used to generate corresponding detailed handover information by calling a large language model based solely on the raw data of the key events; The report synthesis module is used to integrate the global summary information and the detailed handover information according to a preset structure to generate a nursing handover report.
[0005] Secondly, a nursing shift handover method based on the fusion of homogeneous and heterogeneous data and a large language model includes the following steps: Step S1: Obtain structured and unstructured data of the same patient from multiple clinical business systems, perform unified data model alignment and semantic alignment on the structured and unstructured data, and construct a panoramic data view of the patient based on a unified timeline; Step S2: Based on the data changes at adjacent time points in the patient panoramic data view, the patient panoramic data view is divided into multiple clinical events, and a directed association relationship is established between the clinical events based on the temporal correlation and semantic similarity between the clinical events. Step S3: For the clinical events, calculate the anomaly score corresponding to numerical data and the semantic fluctuation score corresponding to text data respectively, and perform a weighted fusion of the anomaly score and the semantic fluctuation score, and filter out key events from the clinical events based on the fused score. Step S4: Based on all the clinical events, call the large language model to generate a global summary of nursing shift handover information; Step S5: Based solely on the raw data of the key events, call the large language model to generate the corresponding detailed handover information; Step S6: Integrate the global summary information and the detailed handover information according to the preset structure to generate a nursing handover report.
[0006] The present invention has the following beneficial effects: This invention achieves unified data model alignment and semantic alignment of homogeneous but heterogeneous data from multiple clinical business systems, and constructs a panoramic data view of patients based on a unified timeline. This enables structured and unstructured data that were originally scattered across different systems to be organized and accessed within the same semantic space, providing a consistent data foundation for subsequent data analysis and processing.
[0007] This invention divides the patient's panoramic data view into multiple clinical events and establishes directed relationships based on the temporal correlation and semantic similarity between the clinical events. This allows changes in the patient's condition within the shift handover time range to be expressed in the form of an event sequence, thereby avoiding the information fragmentation problem caused by presenting information only in the form of a static data list.
[0008] This invention scores numerical and textual data in clinical events separately and then merges the scores to identify key events that require focused handover from multiple clinical events. This allows nursing handover content to be organized around these key events, reducing interference from irrelevant information in the handover process.
[0009] This invention employs a two-layer generation mechanism based on clinical event sequences, generating global summary information covering the entire handover time range and detailed handover information for key events, enabling nursing handover reports to simultaneously provide an overall overview and event-level details, avoiding information loss or content redundancy caused by generating only a single level of text.
[0010] This invention integrates global summary information and detailed handover information corresponding to key events in a structured manner to form a unified nursing handover report. This makes the presentation structure of handover information clearer, making it easier for nursing staff on different shifts to quickly understand the patient's current condition and matters that need continuous attention. Attached Figure Description
[0011] Figure 1 This is a schematic diagram of the nursing shift handover system of the present invention. Detailed Implementation
[0012] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0013] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The first embodiment of the nursing shift handover system based on homogeneous heterogeneous data fusion and large language model in this invention includes: The data fusion module is used to obtain structured and unstructured data of the same patient from multiple clinical business systems, perform unified data model alignment and semantic alignment on the structured and unstructured data, and construct a panoramic data view of the patient based on a unified timeline. The event sequence modeling module is used to segment the patient panoramic data view into multiple clinical events based on the data changes at adjacent time points in the patient panoramic data view, and to construct the logical association strength of the clinical events based on the temporal correlation and semantic similarity between the clinical events. The critical event screening module is used to calculate the anomaly score of numerical data and the semantic fluctuation score of text data for the clinical events, respectively, and to perform a weighted fusion of the anomaly score and the semantic fluctuation score to screen out critical events from the clinical events according to the fusion score. The shift handover generation module includes a first generation unit and a second generation unit, wherein: The first generation unit is used to generate a global summary of nursing handover information by calling a large language model based on all the clinical events. The second generation unit is used to generate corresponding detailed handover information by calling a large language model based solely on the raw data of the key events; The report synthesis module is used to integrate the global summary information and the detailed handover information according to a preset structure to generate a nursing handover report.
[0014] Knowledge fusion of homogeneous and heterogeneous data: By unifying and semantically aligning EMR / LIS / PACS / nursing records / equipment data, a panoramic view of patient data is formed; cross-system event serialization representation is provided, and data is organized for the task of "shift handover".
[0015] For homogeneous but heterogeneous data, it is not just a simple field mapping, but rather the fragmentation of patient behaviors generated in different systems (HIS / LIS / PACS / CIS) and their reorganization into a knowledge sequence with time as the axis and clinical logic as the link.
[0016] The patient panoramic view vector Vpatient is defined as a weighted nonlinear fusion of multi-source data. Through encoding, heterogeneous data of different dimensions are mapped to the same semantic space.
[0017] (1) Multimodal data cleaning and standardization: 1) Structured data (D) struct For indicators such as heart rate, blood pressure, and blood oxygen, the system first filters out outliers. A sliding window W is set (e.g., W=5 minutes), and the average value within the window is calculated. and standard deviation Eliminate intervals Noise beyond that. Then, max-min normalization is used to map all metrics to the [0,1] interval.
[0018] 2) Unstructured data (D) text For nursing record text content, regular expressions are used to remove redundant format symbols, and a pre-trained medical entity recognition model is used to extract symptoms, treatment actions, and negative word entities.
[0019] (2) Tensor construction based on time window This system constructs a unified time axis T. axis D struct and D text Integration.
[0020] 1) Set the minimum time granularity
[0021] 2) For each time step t i Construct fused feature vectors
[0022]
[0023] in, It is a numerical feature extraction layer (consisting of a fully connected layer). This is a text embedding layer.
[0024] 3) If at a certain time step t i If the text record is missing, then Set it to the zero vector; if a value is missing, use spline interpolation to complete it.
[0025] 4) The final generated panoramic view of the patient is represented in the system as a shape of... The dimensional tensor is stored in the vector pool for later use.
[0026] Temporary vector cache: Using patient ID + shift (e.g., P hospital number_NightShift) as the index, the above logical chain is stored in the memory vector pool for immediate use by subsequent LLM inference.
[0027] 2. Event sequence modeling and logical association calculation: Transform discrete time-point data into chains of clinical events with causal implications.
[0028] (1) Segmentation of clinical events: System monitoring The rate of change. When the cosine similarity between the feature vector at a certain moment and the previous moment is lower than a preset threshold. At that time, it was determined to be a new clinical event. The beginning.
[0029] (2) Construction of logical association graph To enable the large language model to understand the causal relationships between events, the system calculates the strength of the logical connections between events.
[0030]
[0031] in, This represents the time difference between two clinical events, and CosineSim represents the cosine similarity between the feature vectors corresponding to the two clinical events. , Indicates the weight.
[0032] Example Parameter Settings: Time weight Setting it to 0.4 reflects the effect of temporal proximity; semantic weight. Set to 0.6, focusing on content similarity; This is the high-dimensional feature vector corresponding to the event.
[0033] Based on the calculated R value, the system constructs a directed weighted graph, retaining only the calculated R value. This step effectively filters out noisy events that are close in time but unrelated to clinical logic, such as pressure injury skin care performed on a patient at the same time and an ENT consultation that may have only a weak correlation.
[0034] 3. Two-tier shift handover generation mechanism and focused scoring calculation This addresses the issue of general large models easily producing illusions or omitting key values.
[0035] (1) Critical event screening (focused scoring) The system processes each event in the sequence. Calculate the score .
[0036] 1) Physical anomaly normalization: Acquiring events The set of offsets for all detection metrics Calculate its L2 norm as the original outlier. Perform a Softmax operation on the A value of all events within the current shift handover cycle. This ensures that regardless of the severity of the patient's underlying condition, Softmax can highlight the moments with the most drastic relative changes within the current cycle, avoiding false alarms caused by the patient's inherent disease patterns, such as a hypertension patient whose baseline blood pressure value has been consistently outside the normal range.
[0037] 2) Semantic fluctuation score extraction: extracting text fragments corresponding to time. Input the distilled LLM, output the probability of negative fluctuations in the condition. .
[0038] 3) Weighted fusion:
[0039] in Anomaly (Dnum): Abnormal deviations in numerical indicators (such as heart rate, test values).
[0040] Softmax will use Anomaly (D num The original value of ) is converted into a probability distribution between 0 and 1.
[0041] Set threshold ,when At that time, the event is marked as a critical event.
[0042] The core of Softmax is to provide a normalization method that maps physical outliers of different dimensions to the (0,1) interval. The degree of anomaly of the physical indicators can be weighted and summed on the same order of magnitude as the semantic sentiment score, so that the anomaly offset can participate in the subsequent weighted calculation in a relatively probabilistic or proportional form.
[0043] (2) Two-layer generation strategy 1) First generation layer (global summary) Input: A summary vector of the event sequence for the entire period.
[0044] Prompt template configuration: "Based on the following time series data, generate a handover summary of no more than 200 words, focusing on describing the overall trend of vital signs (e.g., 'blood pressure shows a slow downward trend')." Output: A macro-level overview of the disease.
[0045] 2) Second generation layer (detail focus) Input: Raw, detailed data for the selected "critical events", including specific test values and complete original nursing records.
[0046] Prompt template configuration: "For the key events that occurred at [time point], please list in detail: 1. Abnormal values; 2. Responsibility measures already taken; 3. Observation items that need to be focused on in the next shift." Output: Structured, detailed event cards.
[0047] (3) Report compilation The system uses rule templates to place the "global summary" at the top of the report and arranges the "detailed event cards" in chronological order in the body of the report, generating the final nursing handover report.
[0048] This invention also relates to a nursing shift handover method, comprising the following steps: Step S1: Obtain structured and unstructured data of the same patient from multiple clinical business systems, perform unified data model alignment and semantic alignment on the structured and unstructured data, and construct a panoramic data view of the patient based on a unified timeline; Step S2: Based on the data changes at adjacent time points in the patient panoramic data view, the patient panoramic data view is divided into multiple clinical events, and a directed association relationship is established between the clinical events based on the temporal correlation and semantic similarity between the clinical events. Step S3: For the clinical events, calculate the anomaly score corresponding to numerical data and the semantic fluctuation score corresponding to text data respectively, and perform a weighted fusion of the anomaly score and the semantic fluctuation score, and filter out key events from the clinical events based on the fused score. Step S4: Based on all the clinical events, call the large language model to generate a global summary of nursing shift handover information; Step S5: Based solely on the raw data of the key events, call the large language model to generate the corresponding detailed handover information; Step S6: Integrate the global summary information and the detailed handover information according to the preset structure to generate a nursing handover report.
[0049] Step S1 includes the following sub-steps: S11. Collect structured and unstructured data of the same patient generated within a preset shift handover time range from multiple clinical business systems, wherein the structured data includes vital sign data, laboratory test data and equipment monitoring data, and the unstructured data includes nursing record text and medical document text. S12. Perform field standardization processing on the structured data so that data items representing the same clinical meaning from different source systems correspond to unified data fields; S13. Perform text preprocessing on the unstructured data, extract semantic units related to the patient's condition and nursing operations, and map the semantic units to a semantic space consistent with the structured data; S14. Based on a unified timeline, perform time alignment on the structured data and the unstructured data, and associate multi-source data within the same time period with the corresponding time nodes; S15. Using time nodes as indexes, merge the structured data and the unstructured data to construct a panoramic data view of the patient.
[0050] Step S2 includes the following sub-steps: S21. Traverse the data features of adjacent time nodes in the panoramic data view in chronological order, and calculate the data difference between adjacent time nodes; S22. When the data difference meets the preset change conditions, the corresponding time node is marked as the new clinical event start point, thereby dividing the panoramic data view into multiple clinical events. S23. For each clinical event, generate an event feature representation to characterize the clinical event, wherein the event feature representation includes structured data features and unstructured data semantic features corresponding to the clinical event; S24. Calculate the correlation between clinical events based on the semantic similarity between the time interval between occurrences of clinical events and the event feature representations; S25. Based on the aforementioned correlation, establish directed correlations between clinical events to form a clinical event sequence structure.
[0051] Step S3 includes the following sub-steps: S31. For each clinical event, calculate the corresponding numerical anomaly score based on the structured data contained in the clinical event. S32. For the same clinical event, calculate the corresponding semantic fluctuation score based on the unstructured data contained in the clinical event; S33. According to the preset fusion rules, the numerical abnormality score and the semantic fluctuation score are weighted and fused to obtain the fusion score of the clinical event. S34. Compare the fusion score with preset screening conditions, and select clinical events that meet the conditions from the clinical events as key events.
[0052] Step S4 includes the following sub-steps: S41. Based on the clinical event sequence structure, extract event summary information covering the entire shift handover time range; S42. Using the event summary information as input, construct input prompts for global summary generation; S43. Call the large language model to process the input prompts and generate global summary information describing the overall changes in the patient's condition.
[0053] Step S5 includes the following sub-steps: S51. For each of the key events, obtain the original structured data and unstructured data corresponding to that key event; S52. Organize the raw data of the key events into event input content according to a preset format; S53. Call the large language model to process the event input content and generate detailed handover information corresponding to the key event.
[0054] Step S6 includes the following sub-steps: S61. According to the preset handover report structure, place the global summary information in the summary section of the report; S62. Arrange the detailed handover information corresponding to the key events in the main body of the report according to the chronological order of the clinical events; S63. Integrate the summary and main text to generate a complete nursing handover report.
[0055] The invention will be demonstrated using a patient who underwent partial nephrectomy in the urology ward as an example.
[0056] Case Background: Patient: Mr. Zhao, male, 55 years old, with clear cell carcinoma of the left kidney, 6 hours after robot-assisted partial nephrectomy of the left kidney.
[0057] 1. Retrieve patient data from different systems: 1) HIS / CIS (Operating Room System): Captures key information during surgery: "Deep tumor location, close to renal hilar vessels, and difficult intraoperative suturing".
[0058] 2) IoT Information: The monitor data shows that the heart rate fluctuates between 100-120 beats / min. Blood pressure is 95 / 55 mmHg.
[0059] 3) LIS: A complete blood count was taken, showing that hemoglobin (Hb) decreased from 145g / L before the operation to 95g / L.
[0060] 4) EMR: Nurse's written record: "Patient complains of thirst and palpitations; perirenal drainage is bright red with small blood clots." Within the past hour, the negative pressure bulb drainage volume suddenly increased to 180ml.
[0061] 2. Constructing a panoramic view and event serialization The system executes an association algorithm to determine active bleeding through logical chains: 1) Logical association: [High surgical difficulty / close to the main blood vessel] → [Fast heart rate / low blood pressure] → [Severe drop in Hb] → [Thirst / palpitation] [Sudden increase in drainage volume + bright red color].
[0062] 2) Identification conclusion: The system automatically identifies these discrete indicators as active bleeding (high-risk) sequences, rather than normal postoperative absorption heat or incision pain.
[0063] 3. Generate a shift handover report 1) I (identity): Zhao, 6 hours after partial nephrectomy of the left kidney.
[0064] 2) S (Current Status): Suspected postoperative active bleeding (requires emergency treatment).
[0065] 3) B (Background): The suture was made near the renal hilum vessels during the operation, which is a high-risk case for bleeding.
[0066] 4) A (Assessment): Extremely high risk: 1. Drainage volume surges by 180ml within 1 hour and is bright red; 2. Hemoglobin decreases by 50g / L compared to preoperative levels; 3. Vital signs show early signs of hemorrhagic shock (increased heart rate, decreased blood pressure, thirst).
[0067] 5) R (Recommendation): 1. Notify the doctor immediately; 2. Increase the rate of intravenous infusion and prepare for crossmatching and transfusion; 3. Instruct the patient to remain in bed; 4. Closely monitor vital signs every 30 minutes.
[0068] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A nursing shift handover system based on homogeneous and heterogeneous data fusion and a large language model, characterized in that: The system includes: The data fusion module is used to obtain structured and unstructured data of the same patient from multiple clinical business systems, perform unified data model alignment and semantic alignment on the structured and unstructured data, and construct a panoramic data view of the patient based on a unified timeline. The event sequence modeling module is used to segment the patient panoramic data view into multiple clinical events based on the data changes at adjacent time points in the patient panoramic data view, and to construct the logical association strength of the clinical events based on the temporal correlation and semantic similarity between the clinical events. The critical event screening module is used to calculate the anomaly score of numerical data and the semantic fluctuation score of text data for the clinical events, respectively, and to perform a weighted fusion of the anomaly score and the semantic fluctuation score to screen out critical events from the clinical events according to the fusion score. The shift handover generation module includes a first generation unit and a second generation unit, wherein: The first generation unit is used to generate a global summary of nursing handover information by calling a large language model based on all the clinical events. The second generation unit is used to generate corresponding detailed handover information by calling a large language model based solely on the raw data of the key events; The report synthesis module is used to integrate the global summary information and the detailed handover information according to a preset structure to generate a nursing handover report.
2. The system according to claim 1, characterized in that, The logical association strength The calculation formula is: in, This represents the time difference between two clinical events, and CosineSim represents the cosine similarity between the feature vectors corresponding to the two clinical events. , Indicates the weight.
3. The system according to claim 1, characterized in that, The fusion score The calculation formula is: in Anomaly(Dnum): Abnormal offset of numerical indicators; Softmax transforms the original values of Anomaly(Dnum) into a probability distribution between 0 and 1; Text fragments corresponding to time Input the distilled LLM, output the probability of negative fluctuations in the condition. .
4. A nursing shift handover method based on homogeneous heterogeneous data fusion and a large language model, characterized in that, Includes the following steps: Step S1: Obtain structured and unstructured data of the same patient from multiple clinical business systems, perform unified data model alignment and semantic alignment on the structured and unstructured data, and construct a panoramic data view of the patient based on a unified timeline; Step S2: Based on the data changes at adjacent time points in the patient panoramic data view, the patient panoramic data view is divided into multiple clinical events, and a directed association relationship is established between the clinical events based on the temporal correlation and semantic similarity between the clinical events. Step S3: For the clinical events, calculate the anomaly score corresponding to numerical data and the semantic fluctuation score corresponding to text data respectively, and perform a weighted fusion of the anomaly score and the semantic fluctuation score, and filter out key events from the clinical events based on the fused score. Step S4: Based on all the clinical events, call the large language model to generate a global summary of nursing shift handover information; Step S5: Based solely on the raw data of the key events, call the large language model to generate the corresponding detailed handover information; Step S6: Integrate the global summary information and the detailed handover information according to the preset structure to generate a nursing handover report.
5. The method according to claim 4, characterized in that, Step S1 includes the following sub-steps: S11. Collect structured and unstructured data of the same patient generated within a preset shift handover time range from multiple clinical business systems, wherein the structured data includes vital sign data, laboratory test data and equipment monitoring data, and the unstructured data includes nursing record text and medical document text. S12. Perform field standardization processing on the structured data so that data items representing the same clinical meaning from different source systems correspond to unified data fields; S13. Perform text preprocessing on the unstructured data, extract semantic units related to the patient's condition and nursing operations, and map the semantic units to a semantic space consistent with the structured data; S14. Based on a unified timeline, perform time alignment on the structured data and the unstructured data, and associate multi-source data within the same time period with the corresponding time nodes; S15. Using time nodes as indexes, merge the structured data and the unstructured data to construct a panoramic data view of the patient.
6. The method according to claim 4, characterized in that, Step S2 includes the following sub-steps: S21. Traverse the data features of adjacent time nodes in the panoramic data view in chronological order, and calculate the data difference between adjacent time nodes. S22. When the data difference meets the preset change conditions, the corresponding time node is marked as the new clinical event starting point, thereby dividing the panoramic data view into multiple clinical events. S23. For each clinical event, generate an event feature representation to characterize the clinical event, wherein the event feature representation includes structured data features and unstructured data semantic features corresponding to the clinical event; S24. Calculate the correlation between clinical events based on the semantic similarity between the time interval between occurrences of clinical events and the event feature representations; S25. Based on the aforementioned correlation, establish directed correlations between clinical events to form a clinical event sequence structure.
7. The method according to claim 4, characterized in that, Step S3 includes the following sub-steps: S31. For each clinical event, calculate the corresponding numerical anomaly score based on the structured data contained in the clinical event. S32. For the same clinical event, calculate the corresponding semantic fluctuation score based on the unstructured data contained in the clinical event; S33. According to the preset fusion rules, the numerical abnormality score and the semantic fluctuation score are weighted and fused to obtain the fusion score of the clinical event. S34. Compare the fusion score with preset screening conditions, and select clinical events that meet the conditions from the clinical events as key events.
8. The method according to claim 4, characterized in that, Step S4 includes the following sub-steps: S41. Based on the clinical event sequence structure, extract event summary information covering the entire shift handover time range; S42. Using the event summary information as input, construct input prompts for global summary generation; S43. Call the large language model to process the input prompts and generate global summary information describing the overall changes in the patient's condition.
9. The method according to claim 4, characterized in that, Step S5 includes the following sub-steps: S51. For each of the key events, obtain the original structured data and unstructured data corresponding to that key event; S52. Organize the raw data of the key events into event input content according to a preset format; S53. Call the large language model to process the event input content and generate detailed handover information corresponding to the key event.
10. The method according to claim 4, characterized in that, Step S6 includes the following sub-steps: S61. According to the preset handover report structure, place the global summary information in the summary section of the report; S62. Arrange the detailed handover information corresponding to the key events in the main body of the report according to the chronological order of the clinical events; S63. Integrate the summary and main text to generate a complete nursing handover report.
Citation Information
Patent Citations
Intelligent power log analysis system based on knowledge graph and large language model
CN119578520A
Nursing shift information association method and system based on dynamic knowledge graph
CN120636739A
Nurse handover management system and method
CN120656673A
Hospital multi-source heterogeneous data integration management system and method based on EMPI
CN120954605A
Intelligent scheduling, scheduling and handover system based on multi-modal AI and knowledge graph
CN121414116A
Cited By
Intracardiac nursing shift change management system and method
CN122025062A
A heart care handover management system and method
CN122025062B
