Infusion safety dynamic monitoring and risk early warning method and system and medium

Abstract

The invention discloses an infusion safety dynamic monitoring and risk early warning method and system and a medium, and the method comprises the steps: collecting and obtaining multi-source data in real time, and carrying out the fusion processing; the infusion real-time data, the real-time monitoring data and the personalized medical data which are obtained after fusion processing are constructed into structured cue words, the structured cue words are input into a large language model, and a structured risk assessment result is generated; and according to a risk assessment result, executing at least one operation of dynamic control, graded early warning and recording. According to the invention, a closed loop of monitoring-evaluation-decision-control is realized, the problem that data of each device is split and linkage analysis cannot be carried out in the prior art is solved, pre-warning is realized, and the rate of missing report and false report is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infusion monitoring and early warning, and in particular relates to an infusion safety dynamic monitoring and risk early warning method and system and a medium. BACKGROUND

[0002] Intravenous infusion is the most commonly used and basic means in clinical practice, but its process is accompanied by many risks, such as infusion reactions, heart failure, and electrolyte disorders. At present, the mainstream infusion safety monitoring technology mainly includes an independent alarm system and a rule-based linkage system.

[0003] The existing intelligent infusion pump and vital sign monitor are independent systems. The infusion pump has functions such as flow rate control, tube blockage, and completion alarm. The monitor can set fixed thresholds for vital sign parameters (such as blood pressure lower limit) to alarm. The defect is that an "information island" is formed between the systems, and the alarm of the monitor is static and passive, and cannot be associated with the type, dose, and speed of the infusion liquid. For example, when a patient with heart dysfunction is infused, the monitor only alarms when the blood pressure has decreased significantly, and cannot provide early warning or intervene in the infusion speed when the blood pressure shows a downward trend.

[0004] For the rule-based linkage system, an "if-then" rule is generally used for control to attempt to realize the linkage of the infusion pump and the monitor. The defect is that the rule base needs to be exhaustively pre-compiled by experts, which is difficult to maintain, and the rule system lacks understanding and reasoning ability for medical context, and cannot handle complex and personalized clinical situations. For example, the system cannot understand the medical record text information that "the patient is an old COPD patient and has been in a low oxygen state for a long time", and may cause false alarms or missed alarms due to fixed thresholds. SUMMARY

[0005] To solve the above problems, the present application provides an infusion safety dynamic monitoring and risk early warning method, system and medium, which deeply fuses real-time sensor data and static, unstructured electronic medical record text data as input for risk assessment, uses the powerful natural language understanding and logical reasoning ability of LLM to replace the traditional and rigid expert rule base, realizes dynamic risk assessment with context awareness, and directly converts the assessment results into specific device control instructions to form an intelligent decision-making and execution closed loop, solving the problem of data fragmentation and inability to link and analyze in the prior art.

[0006] The present application provides an infusion safety dynamic monitoring and risk early warning method, and the specific technical solutions are as follows: S1: Real-time acquisition of multi-source data and fusion processing; The multi-source data includes infusion real-time data output by an infusion pump, real-time monitoring data output by a monitor, and personalized medical data retrieved from a hospital information system. S2: The infusion real-time data, the real-time monitoring data, and the personalized medical data obtained after the fusion processing are constructed into structured prompt words and input into a large language model to generate a structured risk assessment result S3: At least one of dynamic control, hierarchical early warning, and recording is performed according to the risk assessment result.

[0007] Further, the infusion real-time data includes infusion drug name, concentration, current flow rate, cumulative infusion volume, and remaining infusion time. The real-time monitoring data includes heart rate, blood pressure, blood oxygen saturation, and respiratory rate. The individualized medical data includes structured data and unstructured text data, the structured data at least including age, weight, diagnosis information, and the unstructured text data at least including present illness history, past history, and medical order text.

[0008] Further, the fusion processing includes: Adding a unified clock timestamp to the data output by the infusion pump, the monitor, and the hospital information system, respectively; Converting the structured data into key-value pairs according to a pre-defined JSON format, and adding entity labels to the unstructured data and splicing them into a piece of multi-modal text-value mixed input conforming to the large language model prompt word template; Using a set unique identifier as a primary key to link the three types of data in the same time window to the context of the same target.

[0009] Further, the dynamic control operation is as follows: According to the risk assessment result, if the risk level exceeds a pre-set threshold, a control instruction is automatically generated and sent to the infusion pump through an interface to adjust the infusion speed or suspend the infusion.

[0010] Further, the hierarchical early warning operation is as follows: According to the risk assessment result, hierarchical early warning information corresponding to the risk level is generated and pushed to a set target terminal.

[0011] Further, the recording operation is as follows: The input data, output result, and executed action of the current risk assessment are recorded and stored in a database for model continuous optimization and medical quality traceability.

[0012] The application also discloses an infusion safety dynamic monitoring and risk early warning system, which executes the infusion safety dynamic monitoring and risk early warning method. The data acquisition module acquires real-time infusion data of an infusion pump, real-time monitoring data of a monitor and personalized medical data in a hospital information system in real time. The data processing module is used for fusing the multi-source real-time data acquired by the data acquisition module. The large language model analysis module receives the multi-source fused data and performs risk assessment to generate a structured risk assessment result. The decision execution module generates a control instruction and graded early warning information according to the risk assessment result, and issues the control instruction to the infusion pump and sends the graded early warning information to a target terminal.

[0013] The recording module is used for collecting input data, output results and executed actions of current risk assessment and recording and storing in a local database.

[0014] Further, the control instruction is a flow rate adjustment instruction executable by the infusion pump.

[0015] Further, the execution mode of the control instruction is configured as an automatic execution mode or a manual confirmation mode. In the automatic execution mode, the control instruction is directly issued to the infusion pump after being generated. In the manual confirmation mode, the generated control instruction is sent to the target terminal, and the control instruction is issued to the infusion pump after being confirmed by the target terminal.

[0016] The application also provides a computer storage medium storing an infusion safety dynamic monitoring and risk early warning program.

[0017] The application has the following advantages: The application fuses three types of heterogeneous data, i.e., infusion pump data, monitor data and HIS data, into the same context by using unified clock time stamp, primary key association and JSON-entity label hybrid coding, solves the problem of data fragmentation and inability of linkage analysis in the prior art, and realizes the transformation from post-event alarm to pre-event alarm by using the deep understanding of target medical record information by the LLM and combining real-time monitoring data, and makes the assessment more accurate and greatly reduces the occurrence rate of false positives and false negatives. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a method flowchart of the application.

[0019] Figure 2 is a schematic diagram of a system topology architecture of the present application. DETAILED DESCRIPTION

[0020] In the following description, the technical solutions in the embodiments of the present application are clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0021] In the description of the embodiments of the present application, it should be noted that the indicated position or positional relationship is based on the position or positional relationship shown in the drawings, or the position or positional relationship commonly placed when the product of the present application is used, or the position or positional relationship commonly understood by those skilled in the art, or the position or positional relationship commonly placed when the product of the present application is used, which is only for the convenience of describing the present application and simplifying the description, and is not indicative or implied that the indicated device or element must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second" are only used for differentiation and description, and cannot be understood as indicative or implied relative importance.

[0022] In the description of the embodiments of the present application, it should be noted that, unless otherwise explicitly specified and limited, the terms "set", "connected" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be directly connected, or indirectly connected through an intermediate medium. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0023] Embodiment 1 Embodiment 1 of the present application discloses a safety dynamic monitoring and risk early warning method for infusion, as shown in the figure, the specific process is as follows: Figure 1 S1: Real-time acquisition and fusion processing of multi-source data; The multi-source data includes infusion real-time data output by the infusion pump, real-time monitoring data output by the monitor, and personalized medical data called from the hospital information system; Specifically, the infusion real-time data includes infusion drug name, concentration, current flow rate, cumulative infusion volume and remaining infusion time; The real-time monitoring data includes heart rate, blood pressure, blood oxygen saturation and respiratory rate; The real-time monitoring data includes heart rate, blood pressure, blood oxygen saturation and respiratory rate; The personalized medical data includes structured data and unstructured text data. The structured data includes at least age, weight, and diagnostic information, while the unstructured text data includes at least present medical history, past medical history, and medical orders.

[0024] In a preferred embodiment, the fusion process includes: Add a unified clock timestamp to the data output from the infusion pump, monitor, and hospital information system to eliminate the timing deviation caused by the difference in sampling frequency, and ensure that the "blood pressure decrease trend" and "infusion rate change" seen by the subsequent model are causal events within the same time period; For structured data, it is converted into key-value pairs according to the predefined JSON format; for unstructured text, the present medical history and medical orders are kept in the original natural language, but entity labels (ICD-10, RxNorm codes) are added to realize semantic indexing, and finally they are combined into a multimodal text-numerical mixed input that conforms to the prompt word template of the large language model. Using a unique identifier as the primary key, three types of data within the same time window are linked to the context of the same target, preventing the model from mistakenly treating the infusion parameters of target A and the vital signs of target B as the same instance for inference.

[0025] In this embodiment, the three-layer alignment of "time-format-semantics" is completed through the above process.

[0026] S2: The real-time infusion data, real-time monitoring data, and personalized medical data obtained after fusion processing are constructed into structured prompt words and input into a large language model. The large language model performs contextual understanding, risk reasoning, and risk assessment output to generate a structured risk assessment result that includes at least risk level, risk description, and intervention suggestions. Specifically, Large Language Models (LLMs) can be implemented via cloud API calls or local deployment; The contextual understanding refers to parsing and understanding unstructured electronic medical record texts to extract key medical entities and risk factors (such as "heart function class III", "hypokalemia", "penicillin allergy") related to the current infusion.

[0027] The risk reasoning is based on a comprehensive analysis of current infusion parameters (such as "0.9% sodium chloride injection is being infused at a rate of 100 ml / h"), real-time monitoring data (such as "blood pressure drops from 120 / 80 mmHg to 100 / 65 mmHg within 30 minutes"), and personalized medical data (such as "diagnosis: acute myocardial infarction").

[0028] The risk assessment output includes one or more dimensions of risk assessment results, such as: risk level (e.g., "low risk", "medium risk", "high risk"), risk description (e.g., "there is a risk of exacerbating heart failure due to volume overload"), and specific intervention recommendations (e.g., "it is recommended to reduce the flow rate to 50 ml / h and closely monitor blood pressure").

[0029] S3: Based on the risk assessment results, perform at least one of the following operations: dynamic control, tiered early warning, and recording, and return to step S1; Specifically, the operation of the dynamic control is as follows: Based on the risk assessment results, if the risk level exceeds the preset threshold, a control command will be automatically generated and sent to the infusion pump through the interface to adjust the infusion rate or pause the infusion. The generation of the control commands includes: parsing the natural language intervention suggestions output by the large language model into flow rate adjustment commands that the infusion pump can recognize.

[0030] The operation of the tiered early warning system is as follows: Based on the risk assessment results, a graded early warning message corresponding to the risk level is generated and pushed to the designated target terminals, including terminals such as nurse station terminals and mobile nursing terminals.

[0031] The tiered early warning information includes the causes of the risk and intervention suggestions; In this embodiment, control commands are also generated based on interference suggestions; The tiered early warning information can be voice prompts. Specifically, the generated control commands can be set to be sent to the infusion pump for execution only after confirmation by relevant personnel, or they can be set to be sent to the infusion pump for execution automatically.

[0032] The specific operations for recording are as follows: The input data, output results, and actions of the current risk assessment are recorded and stored in the database for continuous model optimization and medical quality traceability.

[0033] Example 2 Embodiment 2 of the present invention discloses an infusion safety dynamic monitoring and risk early warning system based on Embodiment 1 above, such as... Figure 2 As shown, the system includes a data acquisition module, a data processing module, a large language model analysis module, a decision execution module, and a recording module; The data acquisition module acquires real-time infusion data from the infusion pump, real-time monitoring data from the monitor, and personalized medical data from the hospital information system (HIS). Specifically, the infusion pump and the monitor are connected to the bedside terminal deployed on site to transmit data to the bedside terminal in real time; The bedside terminal transmits data to the data acquisition module via a local area network.

[0034] The data processing module is used to fuse and process the multi-source real-time data acquired by the data acquisition module, including: Add a unified clock timestamp to the data output from the infusion pump, monitor, and hospital information system to eliminate the timing deviation caused by the difference in sampling frequency, and ensure that the "blood pressure decrease trend" and "infusion rate change" seen by the subsequent model are causal events within the same time period; For structured data, it is converted into key-value pairs according to the predefined JSON format; for unstructured text, the present medical history and medical orders are kept in the original natural language, but entity labels (ICD-10, RxNorm codes) are added to realize semantic indexing, and finally they are combined into a multimodal text-numerical mixed input that conforms to the prompt word template of the large language model. Using a unique identifier as the primary key, three types of data within the same time window are linked to the context of the same target, preventing the model from mistakenly treating the infusion parameters of target A and the vital signs of target B as the same instance for inference.

[0035] The large language model analysis module receives multi-source fusion data and performs risk assessment to generate structured risk assessment results; The decision execution module generates control instructions and graded early warning information based on the risk assessment results, and sends the control instructions to the infusion pump and the graded early warning information to the target terminal, which is the nurse station monitoring center and the mobile nursing terminal. Specifically, the control command is a flow rate adjustment command that the infusion pump can execute; The execution mode of the control command is configured as either automatic execution mode or manual confirmation mode; In the automatic execution mode, after the control command is generated, it is directly sent to the infusion pump; In the manual confirmation mode, the generated control command is sent to the target terminal, and after confirmation by the target terminal, it is sent to the infusion pump.

[0036] The recording module is used to summarize the input data, output results, and actions performed in the current risk assessment and store them in the local database.

[0037] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. A method for dynamic monitoring and risk early warning of intravenous infusion safety, characterized in that, include: S1: Real-time acquisition and fusion of multi-source data; The multi-source data includes real-time infusion data output by the infusion pump, real-time monitoring data output by the monitor, and personalized medical data retrieved from the hospital information system. S2: The real-time infusion data, real-time monitoring data, and personalized medical data obtained after fusion processing are used to construct structured prompt words and input into a large language model to generate structured risk assessment results; S3: Based on the risk assessment results, perform at least one of the following actions: dynamic control, tiered early warning, and recording.

2. The method for dynamic monitoring and risk warning of infusion safety according to claim 1, characterized in that, The real-time infusion data includes the name of the infused drug, its concentration, current flow rate, cumulative infusion volume, and remaining infusion time. The real-time monitoring data includes heart rate, blood pressure, blood oxygen saturation, and respiratory rate; The personalized medical data includes structured data and unstructured text data.

3. The method for dynamic monitoring and risk warning of infusion safety according to claim 2, characterized in that, The fusion process includes: Add a unified clock timestamp to the data output from infusion pumps, monitors, and the hospital information system; For structured data, it is converted into key-value pairs according to a predefined JSON format; for unstructured data, entity labels are added and the data is combined into a multimodal text-numerical mixed input that conforms to the prompt word template of the large language model. Using a unique identifier as the primary key, multi-source data within the same time window are linked to the context of the same target.

4. The method for dynamic monitoring and risk warning of infusion safety according to claim 1, characterized in that, The dynamic control operation is as follows: Based on the risk assessment results, if the risk level exceeds the preset threshold, a control command is generated and sent to the infusion pump through the interface to adjust the infusion rate or suspend the infusion.

5. The method for dynamic monitoring and risk warning of infusion safety according to claim 1, characterized in that, The operation of the tiered early warning system is as follows: Based on the risk assessment results, tiered early warning information corresponding to the risk level is generated and pushed to the designated target terminals.

6. The method for dynamic monitoring and risk warning of infusion safety according to claim 1, characterized in that, The specific operations for recording are as follows: The input data, output results, and actions taken in the current risk assessment are recorded and stored in the database.

7. A dynamic monitoring and risk early warning system for infusion safety, characterized in that, The method for dynamic monitoring and risk warning of infusion safety according to any one of claims 1-6 includes a data acquisition module, a data processing module, a large language model analysis module, a decision execution module, and a recording module. The data acquisition module acquires real-time infusion data from the infusion pump, real-time monitoring data from the monitor, and personalized medical data from the hospital information system. The data processing module is used to fuse and process the multi-source real-time data acquired by the data acquisition module; The large language model analysis module receives multi-source fusion data and performs risk assessment to generate structured risk assessment results; The decision execution module generates control commands and graded early warning information based on the risk assessment results, and sends the control commands to the infusion pump and the graded early warning information to the target terminal. The recording module summarizes and obtains the input data, output results, and execution actions of the current risk assessment, and records and stores them in the local database.

8. The infusion safety dynamic monitoring and risk early warning system according to claim 7, characterized in that, The control command is a flow rate adjustment command that the infusion pump can execute.

9. The infusion safety dynamic monitoring and risk early warning system according to claim 8, characterized in that, The execution mode of the control command is configured as either automatic execution mode or manual confirmation mode; In the automatic execution mode, after the control command is generated, it is directly sent to the infusion pump; In the manual confirmation mode, the generated control command is sent to the target terminal, and after confirmation by the target terminal, it is sent to the infusion pump.

10. A computer storage medium, characterized in that, When the infusion safety dynamic monitoring and risk warning program, which stores the infusion safety dynamic monitoring and risk warning program, is executed by the processor, it implements the steps of the infusion safety dynamic monitoring and risk warning method according to any one of claims 1-6.

Images