Hospital patient safety management system and method based on data analysis

Abstract

The invention relates to the field of hospital patient safety management, and particularly discloses a hospital patient safety management system and method based on data analysis, and the method comprises the steps: firstly obtaining the hospitalization record of a patient, the heart rate values, collected by an intelligent device, of the patient at a plurality of preset time points, and the blood pressure values, collected by the intelligent device, of the patient at a plurality of preset time points; and then feature extraction and correlation analysis are carried out on the three by using a deep learning technology, and finally whether a patient safety early warning is sent out is judged through a classifier, so that the patient nursing quality is improved, and the response speed to the health condition of the patient is improved.

Description

Technical Field

[0001] The present application relates to the field of hospital patient safety management, and more specifically, to a hospital patient safety management system and method based on data analysis. Background Art

[0002] Patient safety management remains a core challenge in the daily operations of healthcare institutions. A patient's physiological state is dynamic and complex. Abnormal fluctuations in vital signs (such as heart rate and blood pressure) can quickly lead to serious complications, especially for critically ill patients, those recovering from surgery, or those with chronic diseases.

[0003] However, traditional patient safety management systems mostly rely on manual monitoring and recording methods, such as nursing staff manually recording vital signs (such as heart rate and blood pressure) at regular intervals. This results in discrete and delayed data collection, which is unable to capture the patient's real-time dynamic changes, especially when the patient's condition suddenly worsens, and the golden intervention window may be missed. Moreover, in traditional patient safety management systems, manual data entry is susceptible to human errors, such as omissions in records, misreading of values, or deviations caused by fatigue, which affect data reliability. In addition, during the data analysis process, traditional patient safety management systems lack the ability to conduct collaborative analysis of multiple parameters (such as the correlation between abnormal heart rate and blood pressure fluctuations). Traditional patient safety management systems rely only on a single threshold judgment (for example, an alarm is triggered when the heart rate exceeds a fixed value), focusing only on the patient's real-time physiological data, ignoring the complexity of individual differences in patients, historical health status, and the collaborative changes of multiple physiological parameters, making it difficult to accurately identify risks, resulting in a rigid early warning mechanism.

[0004] Therefore, a hospital patient safety management system and method based on data analysis is desired. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a hospital patient safety management system and method based on data analysis. The system first obtains the patient's hospitalization records, the patient's heart rate values at multiple predetermined time points collected by a smart device, and the patient's blood pressure values at multiple predetermined time points collected by a smart device. Then, it uses deep learning technology to extract features and perform correlation analysis on the three. Finally, it uses a classifier to determine whether to issue a patient safety warning, thereby improving the quality of patient care and increasing the speed of response to the patient's health status.

[0006] According to one aspect of the present application, a hospital patient safety management system based on data analysis is provided, which includes:

[0007] A patient safety data acquisition module is used to obtain the patient's hospitalization records, the patient's heart rate values collected by the smart device at multiple predetermined time points, and the patient's blood pressure values collected by the smart device at multiple predetermined time points;

[0008] a patient safety data extraction module, configured to extract a patient hospitalization record-related feature vector and a patient status multimodal feature vector from the patient hospitalization record, the patient's heart rate values at multiple predetermined time points collected by the smart device, and the patient's blood pressure values at multiple predetermined time points collected by the smart device;

[0009] The patient safety warning judgment module is used to judge whether to issue a patient safety warning based on the patient hospitalization record association feature vector and the patient status multimodal feature vector.

[0010] According to another aspect of the present application, a hospital patient safety management method based on data analysis is provided, which includes:

[0011] Obtaining the patient's hospitalization records, the patient's heart rate values collected by the smart device at multiple predetermined time points, and the patient's blood pressure values collected by the smart device at multiple predetermined time points;

[0012] Extracting a patient hospitalization record-related feature vector and a patient state multimodal feature vector from the patient hospitalization record, the patient's heart rate values at multiple predetermined time points collected by the smart device, and the patient's blood pressure values at multiple predetermined time points collected by the smart device;

[0013] Based on the patient hospitalization record associated feature vector and the patient status multimodal feature vector, it is determined whether to issue a patient safety warning.

[0014] Compared with the existing technology, the present application provides a hospital patient safety management system and method based on data analysis, which first obtains the patient's hospitalization records, the patient's heart rate values at multiple predetermined time points collected by smart devices, and the patient's blood pressure values at multiple predetermined time points collected by smart devices, and then uses deep learning technology to perform feature extraction and correlation analysis on the three. Finally, a classifier is used to determine whether to issue a patient safety warning, thereby improving the quality of patient care and increasing the response speed to the patient's health status. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 It is a block diagram of a hospital patient safety management system based on data analysis according to an embodiment of the present application.

[0017] Figure 2 Schematic diagram of data flow of a hospital patient safety management system based on data analysis according to an embodiment of the present application.

[0018] Figure 3 Schematic diagram of a block diagram of a patient safety data extraction module in a hospital patient safety management system based on data analysis according to an embodiment of the present application.

[0019] Figure 4 Schematic diagram of a block diagram of a patient hospitalization record feature extraction unit in a hospital patient safety management system based on data analysis according to an embodiment of the present application.

[0020] Figure 5 This is a block diagram of a patient safety warning judgment module in a hospital patient safety management system based on data analysis according to an embodiment of the present application.

[0021] Figure 6 This is a flowchart of a hospital patient safety management method based on data analysis according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0023] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0024] In addition, numerous specific details are provided in the following detailed description to better illustrate the present application. Those skilled in the art will appreciate that the present application can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present application.

[0025] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0026] Figure 1 This is a block diagram of a hospital patient safety management system based on data analysis according to an embodiment of the present application. Figure 2FIG. 1 is a data flow diagram of a hospital patient safety management system based on data analysis according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the hospital patient safety management system 100 based on data analysis according to an embodiment of the present application includes: a patient safety data acquisition module 110, which is used to obtain patient hospitalization records, heart rate values of patients at multiple predetermined time points collected by smart devices, and blood pressure values of patients at multiple predetermined time points collected by smart devices; a patient safety data extraction module 120, which is used to extract patient hospitalization record-related feature vectors and patient status multimodal feature vectors from the patient hospitalization records, the heart rate values of patients at multiple predetermined time points collected by smart devices, and the blood pressure values of patients at multiple predetermined time points collected by smart devices; a patient safety warning judgment module 130, which is used to judge whether to issue a patient safety warning based on the patient hospitalization record-related feature vectors and the patient status multimodal feature vectors.

[0027] In the aforementioned data analysis-based hospital patient safety management system 100, the patient safety data acquisition module 110 is used to acquire patient hospitalization records, heart rate values collected by smart devices at multiple predetermined time points, and blood pressure values collected by smart devices at multiple predetermined time points. It should be understood that patient safety management is a core challenge in the daily operations of medical institutions, especially for patients with critical illnesses, postoperative conditions, or chronic diseases, where abnormal fluctuations in vital signs can quickly lead to serious complications. However, traditional management systems rely on manual, timed recording of data such as heart rate and blood pressure. This approach not only results in discrete and delayed data collection, making it difficult to capture real-time patient changes, but also easily impacts data reliability due to human error. Furthermore, traditional systems lack the ability to perform multi-parameter collaborative analysis and only use a single threshold for risk warning. This ignores individual patient differences and historical health status, making it impossible to accurately identify risks and resulting in a rigid warning mechanism. Therefore, traditional patient safety monitoring methods face limitations in both timeliness and accuracy. Therefore, the present application processes patient safety data by acquiring patient hospitalization records, heart rate values collected by smart devices at multiple predetermined time points, and blood pressure values collected by smart devices at multiple predetermined time points.

[0028] Specifically, the purpose of obtaining patient hospitalization records and heart rate and blood pressure values collected by smart devices is to comprehensively assess the patient's health status and thus improve patient safety management. Hospitalization records provide key information such as the patient's medical history, treatment plan, and medication usage. This information helps understand the patient's basic health background and potential risk factors. Heart rate and blood pressure values collected by smart devices provide real-time physiological data that reflects the current state of the patient's cardiovascular system. Heart rate and blood pressure are important indicators for assessing heart health and blood circulation. Changes in these data can indicate possible acute worsening of a patient's condition or health issues. Combining hospitalization records with real-time physiological data can create a more comprehensive patient health profile, enabling the safety management system to conduct more accurate risk assessments and issue timely warnings. This allows doctors and caregivers to quickly identify high-risk patients and implement appropriate interventions to prevent potential medical accidents or health issues, thereby improving overall patient safety and treatment outcomes.

[0029] In the aforementioned data analysis-based hospital patient safety management system 100, the patient safety data extraction module 120 is configured to extract patient hospitalization record-related feature vectors and patient status multimodal feature vectors from the patient's hospitalization records, the patient's heart rate values collected by the smart device at multiple predetermined time points, and the patient's blood pressure values collected by the smart device at multiple predetermined time points. It should be understood that by extracting the patient hospitalization record-related feature vectors and the patient status multimodal feature vectors, the system can comprehensively consider historical records and real-time physiological data. This comprehensive analysis approach enables the safety management system to obtain a comprehensive health view and accurately assess a patient's health risks. The extraction and fusion of feature vectors not only improves the accuracy of data analysis but also enables the early warning system to promptly identify abnormalities and issue alerts, supporting the medical team in rapid and effective intervention, optimizing patient safety management, and reducing medical accidents and health risks. It is worth noting that during the extraction of patient hospitalization records, the patient's heart rate values collected by the smart device at multiple predetermined time points, and the patient's blood pressure values collected by the smart device at multiple predetermined time points, data preprocessing steps such as cleaning, deduplication, missing value processing, and outlier detection and processing are required to ensure data quality and accuracy. At the same time, data may need to be standardized or normalized to facilitate subsequent analysis and modeling. Data cleaning removes noise and irrelevant information from the raw data, ensuring the validity of subsequent analysis. This includes, but is not limited to, deduplication, identifying and removing duplicate data entries to avoid bias. Missing value handling addresses gaps or unrecorded data in the dataset. Various strategies can be employed to address missing values, such as directly deleting records with missing values (when the missing percentage is small), using statistical methods (such as mean and median imputation), or applying more complex predictive models to estimate missing values. Outlier detection and handling are also essential. Outliers are data points that deviate from the normal range. They may be caused by measurement errors, data entry errors, or other abnormal factors. If left unaddressed, these outliers can seriously affect the accuracy of analytical results. Therefore, statistical methods (such as boxplots and Z-scores) or machine learning algorithms (such as Isolation Forest and DBSCAN) are needed to identify and handle outliers. These methods may include correcting outliers, replacing them with reasonable estimates, or simply removing them. Furthermore, to facilitate subsequent analysis and modeling, data standardization or normalization is often required. This is because data from different sources often have different dimensions and scales. Unprocessed data can cause certain features to overweight during model training, thus affecting model performance. Standardization typically converts data to zero mean and unit variance, while normalization scales data to a specific range, such as [0, 1].Comprehensive data preprocessing of patient hospitalization records and physiological monitoring data can significantly improve data quality and consistency, providing a solid foundation for subsequent data analysis and modeling. This step not only helps identify potential problems but also ensures that the data used truly reflects the patient's health status, thereby supporting more accurate medical decisions.

[0030] Figure 3 FIG is a block diagram of a patient safety data extraction module in a hospital patient safety management system based on data analysis according to an embodiment of the present application. Figure 3 As shown, in a specific embodiment of the present application, the patient safety data extraction module 120 includes: a patient hospitalization record feature extraction unit 121, used to perform feature extraction on the patient hospitalization record to obtain the patient hospitalization record associated feature vector; a patient heart rate feature extraction unit 122, used to perform feature extraction on the patient's heart rate values at multiple predetermined time points collected by the smart device to obtain the patient's heart rate feature vector; a patient blood pressure feature extraction unit 123, used to perform feature extraction on the patient's blood pressure values at multiple predetermined time points collected by the smart device to obtain the patient's blood pressure feature vector; a patient state multimodal feature association unit 124, used to associate the patient's heart rate feature vector and the patient's blood pressure feature vector to obtain the patient's state multimodal feature vector.

[0031] It should be understood that hospitalization records contain a large amount of unstructured information, such as diagnosis results, treatment plans, medication usage, medical history descriptions, etc., which is crucial for a comprehensive understanding of the patient's health status. The feature extraction process involves converting the original medical records into a series of key features that can capture the patient's health background and potential risks. For example, the diagnosis information in the medical record can be encoded as disease category features, and the treatment plan and medication records can be extracted as treatment pattern features. In this way, the system can use these feature vectors to describe the patient's health status and combine them with real-time physiological data. This feature extraction not only helps to simplify and standardize the data processing process, but also reveals the underlying patterns and trends of the patient's health to further identify abnormal patterns, predict health risks, and generate early warning information. Specifically, this application processes patient hospitalization records through hospitalization record word embedding coding, patient hospitalization record semantic coding model, and patient hospitalization record association convolutional coding model. Among them, the hospitalization record word embedding coding technology converts unstructured text information into a vector representation with specific semantic meaning. This process not only retains the key information in the original text, but also captures the semantic similarity and contextual relationship between words. Building on this foundation, the patient hospitalization record semantic encoding model further processes these word embedding vectors, leveraging advanced natural language processing techniques such as the transformer-based BERT model to implement global, contextual semantic encoding. This encoding approach considers not only the meaning of individual words, but also the role and meaning of words within the entire sentence and even the document. The patient hospitalization record association convolutional encoding model integrates the multiple hospitalization record feature vectors obtained in the above steps, applying convolutional neural network (CNN) mechanisms to identify and extract local dependencies and patterns between hospitalization records, thereby capturing the complex interactions between different records. In this way, the resulting patient hospitalization record association feature vector not only reflects the information of a single record, but also incorporates the interactions and dependencies between records throughout the entire hospitalization period. This comprehensive feature representation provides a comprehensive perspective on the patient's health status, helping to improve the accuracy of disease prediction and enhance risk assessment capabilities.

[0032] Furthermore, feature extraction is performed on the patient's heart rate values collected by the smart device at multiple predetermined time points to obtain a patient heart rate feature vector. The main purpose is to convert dynamic heart rate data into structured information with analytical value, thereby more accurately assessing the patient's heart health. The heart rate data provided by the smart device is typically time series data, recording the patient's heart rate changes at multiple predetermined time points. These changes can reflect the patient's cardiac function, stress state, and potential health issues. The feature extraction process involves converting this raw heart rate data into feature vectors using a specific algorithm. These feature vectors can effectively represent the key characteristics of the heart rate data. First, feature extraction can capture basic statistical properties of the heart rate, such as the mean, standard deviation, maximum, and minimum values. These data can reveal the overall trend and fluctuation range of the heart rate. Second, time series analysis techniques can identify patterns in heart rate changes, such as periodic fluctuations and sudden heart rate fluctuations. These patterns can provide detailed information about heart health, such as the risk of arrhythmia or other heart diseases. Specifically, this application constructs a patient heart rate input vector and processes the patient's heart rate values at multiple predetermined time points using a patient heart rate time series encoder. Constructing the patient's heart rate input vector ensures that heart rate variations at different time points are fully preserved, providing rich context for subsequent analysis. The patient's heart rate time series encoder identifies and captures the temporal dependencies and inherent patterns in the heart rate data, generating a comprehensive feature representation, the patient's heart rate feature vector. This feature vector not only captures the heart rate value at a single time point but also reflects the dynamic variation pattern of the entire time series.

[0033] Furthermore, feature extraction is performed on the patient's blood pressure values collected by a smart device at multiple predetermined time points. The goal is to convert continuous blood pressure data into useful feature vectors to facilitate more accurate assessment and monitoring of the patient's blood pressure health status. Blood pressure data is typically recorded in the form of a time series, covering systolic and diastolic blood pressure values at different time points. This data can reflect the patient's cardiovascular health status and its changing trends, but the raw data may be too complex to be directly analyzed. Feature extraction converts the raw blood pressure data into structured feature vectors, making the data more concise, easier to process, and analyze. First, feature extraction can extract basic blood pressure statistics, such as the mean, standard deviation, maximum, and minimum values. These statistics provide overall characteristics of the blood pressure data and help identify normal blood pressure ranges and fluctuations. Second, by analyzing the time series characteristics of blood pressure data, abnormal fluctuations, trend changes, or cyclical patterns can be identified. These characteristics may reveal potential health issues, such as hypertension, hypotension, or other cardiovascular diseases. Specifically, this application constructs a patient's blood pressure input vector and processes the patient's blood pressure values at multiple predetermined time points using a patient blood pressure time series convolutional neural network model. Observing blood pressure trends over consecutive time periods can reveal a patient's blood pressure fluctuation patterns, potential cardiovascular risks, and treatment effectiveness. The convolutional neural network model for patient blood pressure time series learns local features and their interrelationships within the data, better capturing key characteristics such as trends, periodicity, and abnormal fluctuations in blood pressure.

[0034] Specifically, the purpose of correlating a patient's heart rate and blood pressure eigenvectors to generate a multimodal feature vector of the patient's status is to comprehensively consider multiple physiological indicators, thereby providing a more comprehensive and accurate health assessment. This multimodal feature vector combines information from different data sources to reveal a patient's health status and potential risks, better supporting medical decision-making and risk prediction. Heart rate and blood pressure are two key physiological parameters that reflect cardiac function and blood circulation, respectively. The heart rate eigenvector provides detailed information about heart rhythm and heart rate variability, while the blood pressure eigenvector describes the pressure and fluctuations of blood on blood vessel walls. Analyzing either eigenvector separately can provide some health information, but because they are physiologically interrelated, a single data source may not fully describe a patient's health status. By correlating the heart rate and blood pressure eigenvectors to generate a multimodal feature vector of the patient's status, comprehensive information on cardiac function and blood circulation can be integrated. This feature vector captures the interrelationship between the two physiological indicators, thereby more accurately reflecting the patient's overall health status. It is worth noting that this fusion is not a simple numerical addition, but rather an in-depth analysis of the intrinsic connection between the two physiological signals and their impact on overall health. This application uses a weighted averaging method to assign weights to each feature based on its importance. For example, considering that changes in heart rate may be more indicative of acute disease deterioration than blood pressure in some cases, the weight of the heart rate feature vector may be increased accordingly. This approach ensures that the resulting multimodal feature vector of the patient's state highlights the information most critical to health risk assessment.

[0035] Figure 4 FIG. 1 is a block diagram of a patient hospitalization record feature extraction unit in a hospital patient safety management system based on data analysis according to an embodiment of the present application. Figure 4 As shown, in a specific embodiment of the present application, the patient hospitalization record feature extraction unit 121 includes: a patient hospitalization record semantic feature encoding subunit 1211, used to perform semantic feature encoding on the patient hospitalization record to obtain multiple patient hospitalization record feature vectors; a patient hospitalization record convolution encoding subunit 1212, used to pass the multiple patient hospitalization record feature vectors through a patient hospitalization record associated convolution coding model to obtain the patient hospitalization record associated feature vector.

[0036] It should be understood that semantic feature encoding of patient hospitalization records to generate multiple patient hospitalization record feature vectors aims to transform complex textual information into structured, quantifiable features, thereby improving understanding of patient health status and data processing capabilities. Hospitalization records typically contain a wealth of medical information, including diagnoses, treatment plans, medical histories, medication use, and more. This information is often presented in the form of unstructured text. While this form of data contains a wealth of useful information, it is complex and time-consuming to process. Semantic feature encoding effectively captures and represents key information in text by converting text data into vectors with specific semantic meaning. This process typically involves natural language processing techniques such as word embedding, sentence embedding, or topic modeling. Through these techniques, medical terms, symptom descriptions, and treatment details in the records are converted into numerical feature vectors that capture the semantic relationships and contextual information of the text.

[0037] Furthermore, multiple patient hospitalization record feature vectors are processed through a patient hospitalization record association convolutional coding model to generate a patient hospitalization record association feature vector. This aims to capture and integrate the complex relationships between different information in the hospitalization records, thereby providing a richer and more accurate representation of the patient's health status. Hospitalization record information is often interrelated in terms of time and content. Effectively combining these record feature vectors can help reveal a comprehensive picture of the patient's health status. Convolutional coding models, particularly convolutional neural networks (CNNs), excel at processing data with spatial structure. When applied to hospitalization record feature vectors, these models can extract local relationships and patterns between features through convolution operations. By applying convolution operations to hospitalization record feature vectors, the model can identify and integrate these local relationships, generating a comprehensive association feature vector that not only reflects the information of a single record but also captures the interactions and dependencies between records. The association convolutional coding model is also capable of processing complex patterns in time series data, such as the stage-by-stage changes in disease progression or the progressive nature of treatment effects. By performing convolutional coding on the feature vectors of multiple hospitalization records, the model can identify these patterns and integrate them into the association feature vector. This comprehensive feature representation provides a deep understanding of the patient's health status and helps improve the accuracy of disease prediction and management. Specifically, each layer of the patient hospitalization record association convolutional coding model performs convolution processing, mean pooling processing based on the local feature matrix, and nonlinear activation processing on the input data in the forward pass of the layer to output the patient hospitalization record association feature vector from the last layer of the patient hospitalization record association convolutional coding model, wherein the input of the patient hospitalization record association convolutional coding model is the multiple patient hospitalization record feature vectors.

[0038] In a specific embodiment of the present application, the patient hospitalization record semantic feature encoding subunit 1211 includes: encoding the patient hospitalization record through hospitalization record word embedding to obtain multiple patient hospitalization record data items; and passing the multiple patient hospitalization record data items through the patient hospitalization record semantic encoding model to obtain the multiple patient hospitalization record feature vectors.

[0039] As you can understand, hospitalization records typically contain a wealth of medical information, often in the form of natural language text. To effectively utilize this information, it needs to be converted into a format suitable for computation and analysis. Word embedding technology maps each word or phrase in the text into a high-dimensional continuous vector space. Each word's vector not only retains the semantic information of the word but also captures the relationships between words. Through word embedding, medical terms, symptom descriptions, and treatment details in hospitalization records can be converted into semantically meaningful vectors. In this way, each data item in the hospitalization record, such as symptom descriptions, test results, and treatment plans, is represented as a vector. These vectors reflect the semantic information of the text, facilitating further processing and analysis. A key advantage of word embedding is its ability to capture semantic similarity and contextual relationships within vocabulary. For example, similar conditions or treatments are mapped to similar vector positions, enabling models to identify and associate related information. By applying word embedding to all hospitalization records, large amounts of unstructured text data can be converted into structured data items for multiple patients. These data items can include descriptions of medical conditions, treatment processes, medication usage, and more, facilitating the generation of unified patient health records.

[0040] Furthermore, multiple patient hospitalization record data items are processed through a patient hospitalization record semantic encoding model to obtain multiple patient hospitalization record feature vectors, aiming to convert rich and complex medical text data into structured feature representations with high semantic information to facilitate more in-depth data analysis and mining. Hospitalization records usually exist in the form of unstructured text, which is difficult to directly perform quantitative analysis. Semantic encoding models, such as word embedding technology, language models in deep learning (such as BERT, GPT, etc.), or methods based on topic models, can convert each data item in the hospitalization record into a high-dimensional vector. These models not only capture the semantic information of individual words, but also recognize and encode the contextual relationships and implicit semantic structures in the text. In this way, each data item in the hospitalization record is mapped to a semantic feature vector, which can reflect the complex semantic relationships and content information between data items in a high-dimensional space. Specifically, the embedding layer of the patient hospitalization record semantic encoding model is used to map each hospitalization record word in the multiple patient hospitalization record data items into a word embedding vector to obtain a sequence of hospitalization record word embedding vectors; and the converter-based Bert model of the patient hospitalization record semantic encoding model is used to perform global context semantic encoding on the sequence of hospitalization record word embedding vectors to obtain multiple patient hospitalization record feature vectors.

[0041] In a specific embodiment of the present application, the patient heart rate feature extraction unit 122 includes: constructing the patient's heart rate values at multiple predetermined time points collected by the smart device into a patient heart rate input vector; and passing the patient heart rate input vector through a patient heart rate timing encoder to obtain the patient heart rate feature vector.

[0042] It's understandable that smart devices, such as smartwatches or health monitoring devices, can collect real-time heart rate data from patients. This data is typically presented as a time series, consisting of heart rate values recorded at different points in time. This time series of heart rate data captures changes in a patient's heart rate over time, which can reflect their heart health, physical activity level, emotional state, and more. Constructing heart rate values at multiple time points as input vectors preserves the continuity and trend information of the time series, providing rich context for analytical models.

[0043] Furthermore, the patient's heart rate input vector is passed through a patient heart rate time series encoder to obtain a patient heart rate feature vector. This is done to extract deep patterns and trends from time series data, thereby providing more accurate information for health monitoring and disease prediction. Heart rate data collected by smart devices is typically in the form of time series, consisting of heart rate values recorded at different time points. These data points have certain temporal dependencies and sequential relationships, making it difficult to directly process these data to capture the deep characteristics of their dynamic changes. Therefore, a heart rate time series encoder is used to effectively capture and encode the temporal dependencies and inherent patterns of this time series data. Heart rate time series encoders, typically based on models such as recurrent neural networks (RNNs), long short-term memory networks (LSTMs), or gated recurrent units (GRUs), are specifically designed to process and analyze time series data. These encoders can consider the temporal relationships between data points, namely, how the heart rate value at one time point affects the values at subsequent time points and how this is reflected in long-term heart rate trends. In this way, the time series encoder learns the temporal dependencies and long-term memory effects in the data, thereby extracting more representative and informative features. When the heart rate input vector is passed to the time series encoder, the model will gradually process the heart rate value at each time point, capture its relationship with the previous and next time points, and integrate this information to generate a comprehensive feature representation. This feature vector not only contains the heart rate value information at a single time point, but also reflects the dynamic change pattern of the entire time series. This feature vector can reveal key features such as the trend, periodicity, and abnormal fluctuations of heart rate changes, which helps to detect potential health problems such as heart attacks, excessive stress, or arrhythmias. Specifically, the patient heart rate input vector is fully connected using the fully connected layer of the patient heart rate time series encoder to extract high-dimensional implicit features of the eigenvalues at each position in the patient heart rate input vector; and the patient heart rate input vector is one-dimensionally encoded using the one-dimensional convolutional layer of the patient heart rate time series encoder to extract high-dimensional implicit correlation features of the correlation between the eigenvalues at each position in the patient heart rate input vector.

[0044] In a specific embodiment of the present application, the patient blood pressure feature extraction unit 123 includes: constructing the blood pressure values of the patient at multiple predetermined time points collected by the smart device into a patient blood pressure input vector; and passing the patient blood pressure input vector through a patient blood pressure time series convolutional neural network model to obtain the patient blood pressure feature vector.

[0045] It should be understood that smart devices, such as smart watches or blood pressure monitors, are able to record patients' blood pressure values at predetermined time intervals. These data usually exist in the form of time series, including blood pressure measurements at multiple time points. In order to deeply analyze these data and extract useful health information, these time series data need to be organized into a unified input vector. The main purpose of constructing a blood pressure input vector is to integrate the blood pressure values at different time points into a fixed format data representation so that these data can be further processed and analyzed. The blood pressure value in the time series data is not just the data of a single point, but also reflects the patient's blood pressure change trend and fluctuation pattern. By constructing a blood pressure input vector, these patterns and trends can be more easily identified, resulting in a more comprehensive health assessment.

[0046] Furthermore, the patient's blood pressure input vector is processed through a convolutional neural network (CNN) model for the patient's blood pressure time series to obtain a blood pressure feature vector. This is done to leverage the powerful ability of CNNs to capture local features and patterns in time series data, thereby extracting richer and more meaningful information about blood pressure changes. Blood pressure time series data consists of measurements at multiple time points, which exhibit temporal dependencies and local patterns between these data points. Traditional data processing methods can struggle to effectively identify and utilize these details. Convolutional neural networks, through their unique convolution operation, are able to detect local features of varying scales and locations within a time series, giving them a significant advantage in processing blood pressure time series data. By applying convolutional filters to locally scan the time series data, CNNs are able to identify important patterns in the blood pressure data, such as periodic fluctuations, sudden changes, or long-term trends. These convolutional filters are able to capture key features in the time series, extracting meaningful information about health status, and transforming it into a high-dimensional feature representation. Specifically, the patient blood pressure input vector is fully-connected encoded using the fully-connected layer of the patient blood pressure time series convolutional neural network model to extract high-dimensional implicit features of the eigenvalues at each position in the patient blood pressure input vector; and the patient blood pressure input vector is one-dimensionally encoded using the one-dimensional convolutional layer of the patient blood pressure time series convolutional neural network model to extract high-dimensional implicit correlation features of the correlation between the eigenvalues at each position in the patient blood pressure input vector.

[0047] In the aforementioned data analysis-based hospital patient safety management system 100, the patient safety warning determination module 130 is configured to determine whether to issue a patient safety warning based on the patient hospitalization record-related feature vector and the patient status multimodal feature vector. It should be understood that determining whether to issue a patient safety warning based on the patient hospitalization record-related feature vector and the patient status multimodal feature vector is intended to comprehensively utilize data from different sources, improve the ability to monitor patient health status, and promptly identify potential safety risks. The patient hospitalization record-related feature vector typically includes the patient's medical history, treatment plan, and past hospitalization information, which can reflect the patient's health background and potential risk factors. Furthermore, the patient status multimodal feature vector may incorporate real-time monitoring data such as blood pressure, heart rate, and body temperature, as well as other auxiliary information such as laboratory test results and imaging data. By comprehensively considering a patient's real-time physiological data and long-term health records, this multimodal feature vector can provide a more comprehensive and accurate health assessment, more accurately determining whether a safety warning is necessary and preventing potential health crises.

[0048] Figure 5 FIG. 1 is a block diagram of a patient safety early warning judgment module in a hospital patient safety management system based on data analysis according to an embodiment of the present application. Figure 5 As shown, in a specific embodiment of the present application, the patient safety warning judgment module 130 includes: a patient safety data feature fusion unit 131, used to fuse the patient hospitalization record association feature vector and the patient status multimodal feature vector to obtain a patient safety warning judgment feature vector; a patient safety data feature optimization unit 132, used to perform element-by-element entropy association feature regression iterative optimization on the patient safety warning judgment feature vector to obtain an optimized patient safety warning judgment feature vector; a patient safety warning classification judgment unit 133, used to pass the optimized patient safety warning judgment feature vector through a classifier to obtain a classification result, and the classification result is used to determine whether to issue a patient safety warning.

[0049] It should be understood that by fusing the patient hospitalization record-related feature vectors with the patient status multimodal feature vectors to obtain a patient safety warning judgment feature vector, the dynamic changes in patient health and potential risks can be more comprehensively captured. Such a comprehensive analysis helps identify complex health issues and potential danger signals, improving the accuracy and reliability of the early warning system. Specifically, the present application uses a weighted average method to assign different weights to each feature based on its importance. In another specific embodiment of the present application, an attention mechanism is applied to dynamically adjust the impact of different features on the final result, ensuring that key information receives more attention. The attention mechanism calculates the importance weight of each feature by matching a query with a set of key-value pairs. In this process, the query can be regarded as the current piece of information that requires attention, while the key-value pairs represent all potentially relevant feature information. By calculating the similarity (or affinity) between the query and each key, a series of weight scores can be obtained. These weight scores reflect the relevance and importance of each feature to the specific query. Subsequently, these weight scores are used as coefficients to perform a weighted summation of the corresponding values to obtain a reweighted feature representation. This process enables the model to automatically focus on the most relevant information while suppressing less important details. For example, when assessing a patient's health risks, changes in heart rate or blood pressure at certain moments may be more critical than at other moments. The attention mechanism can identify these key moments and give them higher weights so that they dominate the final feature representation. It is worth noting that the attention mechanism is not static, but dynamically adjusts its focus based on the different input data. This means that the focus of the model may vary for different patients or different time points for the same patient. By introducing the attention mechanism, the most critical information can be emphasized, helping the system make more informed decisions, thereby more accurately integrating data from different sources and improving the ability to predict potential health risks.

[0050] In particular, in the technical solution of the present application, feature processing and feature fusion are performed on the patient's hospitalization records, the patient's heart rate values at multiple predetermined time points, and the patient's blood pressure values at multiple predetermined time points to obtain a patient safety warning judgment feature vector. When extracting features from the hospitalization records and physiological data, the hospitalization record word embedding encoding and semantic encoding model may not perfectly capture all details of the patient's condition, resulting in feature errors. Similarly, the design and training of the heart rate time series encoder and blood pressure time series convolutional neural network model may also introduce certain errors. In the data fusion process to obtain the patient safety warning judgment feature vector, a slight perturbation in any feature may affect the generation of the final patient safety warning judgment feature vector. Moreover, the fusion strategy may be sensitive to small-scale feature changes, thereby introducing local perturbations. Considering that the operation of the classifier depends on the quality and diversity of the input patient safety warning judgment feature vector, if the patient safety warning judgment feature vector is unstable due to perturbations during the data processing process, the classifier will have insufficient classification accuracy. Therefore, in the technical solution of the present application, the patient safety warning judgment feature vector is subjected to element-by-element entropy correlation feature regression iterative optimization to obtain an optimized patient safety warning judgment feature vector.

[0051] In a specific embodiment of the present application, the patient safety data feature optimization unit 132 is used to calculate the autocorrelation matrix of the patient safety early warning judgment feature vector, and perform principal component analysis on the autocorrelation matrix of the patient safety early warning judgment feature vector to obtain a set of patient safety early warning judgment feature eigencomponent coding vectors, which can be expressed as follows:

[0052]

[0053] Among them, V represents the patient safety warning judgment feature vector, T represents the transpose of the vector, and M z represents the autocorrelation matrix, U represents the set of encoding vectors of the characteristic eigencomponents of patient safety early warning judgment, v1, v2, v m denote the first, second, and mth patient safety early warning judgment feature eigencomponent encoding vectors, Λ denotes the patient safety early warning judgment feature diagonal matrix after principal component analysis, λ1, λ m They respectively represent the first and mth eigenvalues on the diagonal of the patient safety early warning judgment feature diagonal matrix.

[0054] That is, because abnormal fluctuations in patients' physiological parameters (such as heart rate and blood pressure) are often not isolated events, but rather the coupled manifestation of multi-dimensional features in a time series, this application introduces the calculation of the autocorrelation matrix, which can effectively reveal the interdependence of patient safety warning judgment features in the time dimension, such as whether the periodic fluctuation of a certain parameter is potentially associated with the abnormal trend of another parameter. The autocorrelation matrix is then reduced in dimensionality through principal component analysis (PCA), and the key information of the feature space is reconstructed from the perspective of linear algebra, enabling the model to achieve more accurate risk prediction based on the dynamic association between the patient's historical data features and real-time multimodal parameters.

[0055] In a specific embodiment of the present application, the patient safety data feature optimization unit 132 is further configured to input the set of patient safety early warning judgment feature intrinsic component encoding vectors into a sequence encoder based on a forward LSTM model to obtain a set of patient safety early warning judgment feature intrinsic component context-related encoding vectors, which is expressed as follows:

[0056] F=LSTM([v1,v2,…,v m ])=[s1,s2,…,s m ]

[0057] Among them, LSTM represents the forward LSTM model, F represents the set of context-related encoding vectors of the intrinsic component of the patient safety warning judgment feature, s1, s2, s m Represents the context-associated encoding vector of the first, second, and mth patient safety warning judgment feature intrinsic component.

[0058] That is, through the sequence modeling of LSTM, the model can analyze the implicit temporal dependency of the encoding vector of the patient safety warning judgment feature intrinsic component in a time-recursive manner, making up for the limitations of PCA on nonlinear associations, and transforming the originally static patient safety warning judgment feature intrinsic component into a patient safety warning judgment feature intrinsic component context-related encoding vector with dynamic context semantics, so that subsequent classifiers can achieve more accurate risk prediction based on the temporal evolution laws in the patient's historical data.

[0059] In a specific embodiment of the present application, the patient safety data feature optimization unit 132 is further used to calculate the bit-by-bit displacement entropy between each corresponding patient safety early warning judgment feature intrinsic component context-associated coding vector and the patient safety early warning judgment feature intrinsic component coding vector in the set of the patient safety early warning judgment feature intrinsic component context-associated coding vectors and the patient safety early warning judgment feature intrinsic component coding vectors to obtain a set of bit-by-bit displacement entropies, which is expressed as follows:

[0060]

[0061] Among them, ∧ represents a logical operator, For all conditions that meet The elements are bitwise shifted, w represents the bitwise comparison function, v i represents the encoding vector of the eigencomponent of the i-th patient safety early warning judgment feature, represents the eigenvalue of the jth position of the eigencomponent encoding vector of the i-th patient safety early warning judgment feature, s i represents the context-related encoding vector of the eigencomponent of the i-th patient safety early warning judgment feature, represents the eigenvalue of the jth position of the context-related coding vector of the eigencomponent of the i-th patient safety early warning judgment feature, ε represents the predetermined threshold, r i represents the i-th bitwise shift matching feature vector, represents the eigenvalue of the jth position of the ith bitwise shift matching eigenvector, L represents the length of the ith bitwise shift matching eigenvector, e i represents the i-th bitwise shift entropy.

[0062] That is, this application introduces bit-by-bit displacement entropy, which can capture subtle information disturbances that cannot be identified by traditional measurement methods by analyzing the change pattern of binary bits and information entropy gain. And through information theory-oriented quantitative indicators, the originally abstract binary differences are transformed into interpretable risk-sensitive signals, which not only solves the problem that traditional measurements are insensitive to high-order nonlinear correlations, but also provides a refined decision-making basis based on information entropy changes for subsequent dynamic adjustment of feature weights and multimodal data fusion, ultimately improving the model's recognition accuracy for sudden changes in complex physiological states and avoiding the risk of misjudgment caused by ignoring the directionality of information disturbances at the binary level (such as entropy increase or entropy decrease).

[0063] In a specific embodiment of the present application, the patient safety data feature optimization unit 132 is further configured to perform a weighting process based on a Softmax function on the set of bitwise displacement entropies to obtain a set of bitwise displacement entropy optimization factors, which is expressed as follows:

[0064] a i =softmax(e i )

[0065] Among them, softmax represents the normalized exponential function, a i represents the i-th bitwise shift entropy optimization factor.

[0066] That is, the Softmax function converts entropy values into probability distributions through exponential mapping, so that the intrinsic components corresponding to high entropy values receive significantly higher weights, while low entropy components are suppressed. Specifically, through the smoothness constraint of Softmax, weight distribution not only avoids optimization instability caused by extreme values, but also achieves differentiated competition based on information theory, thereby enhancing the system's sensitivity to complex risk events, converting abstract information entropy changes into interpretable and operational weight parameters, and generating a bit-by-bit displacement entropy optimization factor.

[0067] In a specific embodiment of the present application, the patient safety data feature optimization unit 132 is further configured to: based on the set of bitwise shift entropy optimization factors, fuse the set of patient safety early warning judgment feature intrinsic component encoding vectors to obtain an optimized patient safety early warning judgment feature vector, which is expressed as follows:

[0068]

[0069] Among them, v f Represents the optimized patient safety early warning judgment feature vector.

[0070] That is, by using the bitwise shift entropy optimization factor as a weight coefficient, the model can differentiate and enhance features based on the hierarchy of information entropy gain. Through this fusion method, the originally static feature fusion is transformed into a dynamic information value distribution. This not only solves the problem of insufficient warning sensitivity caused by traditional methods that ignore differences in feature importance, but also improves the model's ability to capture sudden changes in complex physiological states by retaining high information density features. Ultimately, the generated optimized patient safety warning judgment feature vector can more efficiently adapt to the needs of subsequent classification tasks.

[0071] Furthermore, the optimized patient safety warning judgment feature vector is processed by a classifier to obtain a classification result, which is used to determine whether to issue a patient safety warning. This is to transform comprehensive health information into specific decisions, thereby effectively identifying safety risks that require attention. A classifier is a machine learning algorithm that can perform pattern recognition and classification based on the input optimized patient safety warning judgment feature vector to produce corresponding results. In a medical warning system, the classifier analyzes the optimized patient safety warning judgment feature vector and maps complex health data into predefined categories, such as "normal" or "warning," thereby helping medical personnel make quick decisions. The classification result serves as the basis for determining whether to issue a patient safety warning, helping the medical team to quickly take necessary actions, such as adjusting treatment plans or conducting further examinations, to ensure patient safety. In a specific embodiment of the present application, a commonly used classifier is a support vector machine (SVM). SVM uses kernel functions (such as linear, polynomial, or RBF kernels) to find the optimal splitting plane in the case of nonlinear separability to distinguish different safety warning levels. SVM can effectively process data in high-dimensional spaces and is suitable for medical data environments with complex features. It can effectively classify and reduce the risk of overfitting. Another possible option is the random forest classifier, a decision tree-based ensemble learning method. This method constructs multiple decision trees, each trained on a subset of the data, with each node split based on a randomly selected subset of features. Each decision tree then independently predicts the input optimized patient safety warning feature vector, and the predictions of all decision trees are combined (typically using a majority voting mechanism) to determine the final classification result. This method improves accuracy and controls overfitting, making it suitable for processing datasets with large numbers of features and complex interrelationships. It can also effectively assess feature importance, thereby providing more stable and reliable patient safety warning judgments when faced with complex medical data.

[0072] It is worth mentioning that in a specific embodiment of the present application, an independent data set is used for verification and evaluation to test the model's predictive accuracy and generalization ability. Common evaluation indicators include accuracy, recall, F1 score, area under the ROC curve (AUC), etc. Among them, accuracy refers to the ratio of the number of correctly classified samples to the total number of samples, which measures the proportion of correctly classified samples in the model's prediction results. Recall focuses on how many samples that are actually positive are correctly identified, emphasizing the model's ability to discover relevant instances. For medical early warning systems, a high recall rate means that the system can effectively identify patients who are truly facing health risks, thereby reducing the possibility of missed diagnosis. The F1 score is the harmonic mean of accuracy and recall, providing a way to balance the two. It is particularly suitable for cases of class imbalance, ensuring that the model is neither too conservative nor too radical. The area under the ROC curve (AUC) is another important evaluation criterion, which shows the strength of the classifier's discrimination ability. The closer the AUC value is to 1, the stronger the model's discrimination ability. By plotting the true positive rate (i.e., recall rate) against the false positive rate, AUC provides an intuitive way to view the performance of the classifier across the entire range of decision thresholds. This is particularly suitable for evaluating binary classification problems that require operations at different thresholds. Together, these evaluation metrics provide a comprehensive and in-depth understanding of the model, helping to optimize the hospital patient safety management system and improve its reliability and effectiveness in practical applications. Next, optimization is performed based on the validation results to improve the performance of the model. Once the optimization is complete, the model can be deployed in actual application scenarios for risk assessment and early warning of chronic diseases.

[0073] In summary, the embodiment of the present application first obtains the patient's hospitalization record, the patient's heart rate value at multiple predetermined time points collected by the smart device, and the patient's blood pressure value at multiple predetermined time points collected by the smart device, and then uses the hospitalization record word embedding coding technology, the patient's hospitalization record semantic coding model, and the patient's hospitalization record association convolution coding model to process the patient's hospitalization record to extract the semantic association features of the patient's hospitalization record, and then uses the patient's heart rate time series encoder and the patient's blood pressure time series convolutional neural network model to process the patient's heart rate value at multiple predetermined time points and the patient's blood pressure value at multiple predetermined time points, and associate them to obtain the patient's multimodal features associated with the patient's heart rate features and the patient's blood pressure, and then performs association analysis optimization on the patient's multimodal features and the semantic association features of the patient's hospitalization record, and uses a classifier to determine whether to issue a patient safety warning, thereby improving the quality of patient care and improving the response speed to the patient's health status. In this way, the data discrete lag problem caused by the traditional manual recording method is overcome, and the possibility of human error is reduced, ensuring the accuracy and reliability of the data. In addition, through the application of deep learning technology, the system can not only process real-time physiological data, but also conduct comprehensive analysis in combination with the patient's hospitalization records. This allows it to take into account individual differences, historical health status and other factors when assessing the patient's health status, greatly improving the accuracy of risk identification. Compared with the traditional single threshold judgment mechanism, this multi-dimensional information fusion method can better capture the intrinsic connection between different physiological parameters and their coordinated change trends, which helps to discover potential risk patterns and provide early warning of possible health crises. This solution not only takes into account the complexity of the data itself, but also introduces a local shift deviation compensation mechanism to enhance the classifier's robustness to local perturbations in the input feature vector, improve the system's stability and generalization ability, thereby achieving comprehensive monitoring and timely response to the patient's health status, effectively making up for the shortcomings of the traditional safety management system, and providing strong support for improving the quality of patient care and medical services in hospitals.

[0074] As described above, the data analysis-based hospital patient safety management system 100 according to an embodiment of the present application can be implemented in various terminal devices. In one example, the data analysis-based hospital patient safety management system 100 can be integrated into the terminal device as a software module and / or hardware module. For example, the data analysis-based hospital patient safety management system 100 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the data analysis-based hospital patient safety management system 100 can also be one of the many hardware modules of the terminal device.

[0075] Alternatively, in another example, the data analysis-based hospital patient safety management system 100 and the terminal device may also be separate devices, and the data analysis-based hospital patient safety management system 100 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0076] Figure 6 FIG is a flow chart of a hospital patient safety management method based on data analysis according to an embodiment of the present application. Figure 6 As shown, according to the hospital patient safety management method based on data analysis according to the embodiment of the present application, it includes: S110, obtaining the patient's hospitalization records, the patient's heart rate values at multiple predetermined time points collected by the smart device, and the patient's blood pressure values at multiple predetermined time points collected by the smart device; S120, extracting the patient's hospitalization record-related feature vector and the patient's status multimodal feature vector from the patient's hospitalization records, the patient's heart rate values at multiple predetermined time points collected by the smart device, and the patient's blood pressure values at multiple predetermined time points collected by the smart device; S130, judging whether to issue a patient safety warning based on the patient's hospitalization record-related feature vector and the patient's status multimodal feature vector.

[0077] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned hospital patient safety management method based on data analysis have been referred to above. Figures 1 to 5 The present invention has been introduced in detail in the description of the hospital patient safety management system based on data analysis, and therefore, its repeated description will be omitted.

[0078] It should be noted that the information involved in this application (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals (including but not limited to signals transmitted between user terminals and other devices, etc.) are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions. For example, the relevant information or data such as "patient hospitalization records", "patient's heart rate values at multiple predetermined time points", "patient's blood pressure values at multiple predetermined time points" involved in this application are all obtained with full authorization.

[0079] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the module division is merely a logical function division, and other division methods may be used in actual implementation.

[0080] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.

[0081] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0082] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0083] Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the claims are intended to be embraced therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0084] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. Second-order terms are used to indicate names and do not imply any particular order.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit of the technical solutions of the present invention.

Claims

1. A hospital patient safety management system based on data analysis, characterized in that: include: A patient safety data acquisition module is used to obtain the patient's hospitalization records, the patient's heart rate values collected by the smart device at multiple predetermined time points, and the patient's blood pressure values collected by the smart device at multiple predetermined time points; a patient safety data extraction module, configured to extract a patient hospitalization record-related feature vector and a patient status multimodal feature vector from the patient hospitalization record, the patient's heart rate values at multiple predetermined time points collected by the smart device, and the patient's blood pressure values at multiple predetermined time points collected by the smart device; The patient safety warning judgment module is used to judge whether to issue a patient safety warning based on the patient hospitalization record association feature vector and the patient status multimodal feature vector.

2. The hospital patient safety management system based on data analysis according to claim 1 is characterized in that: The patient safety data extraction module includes: A patient hospitalization record feature extraction unit, configured to extract features from the patient hospitalization record to obtain a feature vector associated with the patient hospitalization record; a patient heart rate feature extraction unit, configured to extract features from the patient's heart rate values at a plurality of predetermined time points collected by the smart device to obtain a patient heart rate feature vector; a patient blood pressure feature extraction unit, configured to extract features from the patient's blood pressure values at a plurality of predetermined time points collected by the smart device to obtain a patient blood pressure feature vector; The patient state multimodal feature associating unit is configured to associate the patient heart rate feature vector with the patient blood pressure feature vector to obtain the patient state multimodal feature vector.

3. The hospital patient safety management system based on data analysis according to claim 2 is characterized in that: The patient hospitalization record feature extraction unit includes: a patient hospitalization record semantic feature encoding subunit, configured to perform semantic feature encoding on the patient hospitalization record to obtain a plurality of patient hospitalization record feature vectors; The patient hospitalization record convolutional coding subunit is used to pass the multiple patient hospitalization record feature vectors through the patient hospitalization record associated convolutional coding model to obtain the patient hospitalization record associated feature vector.

4. The hospital patient safety management system based on data analysis according to claim 3 is characterized in that: The patient hospitalization record semantic feature encoding subunit includes: The patient hospitalization record is encoded by hospitalization record word embedding to obtain a plurality of patient hospitalization record data items; The plurality of patient hospitalization record data items are passed through a patient hospitalization record semantic encoding model to obtain the plurality of patient hospitalization record feature vectors.

5. The hospital patient safety management system based on data analysis according to claim 4 is characterized in that: The patient heart rate feature extraction unit includes: constructing the patient's heart rate values at a plurality of predetermined time points collected by the smart device into a patient's heart rate input vector; The patient heart rate input vector is passed through a patient heart rate timing encoder to obtain the patient heart rate feature vector.

6. The hospital patient safety management system based on data analysis according to claim 5 is characterized in that: The patient blood pressure feature extraction unit comprises: constructing the patient's blood pressure values at multiple predetermined time points collected by the smart device into a patient's blood pressure input vector; The patient's blood pressure input vector is passed through a patient's blood pressure time series convolutional neural network model to obtain the patient's blood pressure feature vector.

7. The hospital patient safety management system based on data analysis according to claim 6 is characterized in that: The patient safety early warning judgment module includes: a patient safety data feature fusion unit, configured to fuse the patient hospitalization record associated feature vector and the patient status multimodal feature vector to obtain a patient safety early warning judgment feature vector; a patient safety data feature optimization unit, configured to perform element-by-element entropy correlation feature regression iterative optimization on the patient safety early warning judgment feature vector to obtain an optimized patient safety early warning judgment feature vector; The patient safety warning classification judgment unit is used to pass the optimized patient safety warning judgment feature vector through a classifier to obtain a classification result, and the classification result is used to determine whether to issue a patient safety warning.

8. The hospital patient safety management system based on data analysis according to claim 7 is characterized in that: The patient safety data feature optimization unit is used to: Calculating an autocorrelation matrix of the patient safety early warning judgment feature vector, and performing principal component analysis on the autocorrelation matrix of the patient safety early warning judgment feature vector to obtain a set of patient safety early warning judgment feature eigencomponent coding vectors; Inputting the set of patient safety early warning judgment feature intrinsic component encoding vectors into a sequence encoder based on a forward LSTM model to obtain a set of patient safety early warning judgment feature intrinsic component context-related encoding vectors; Calculating the bit-by-bit displacement entropy between each corresponding group of the patient safety early warning judgment feature intrinsic component context-associated coding vectors and the patient safety early warning judgment feature intrinsic component coding vectors in the set of the patient safety early warning judgment feature intrinsic component coding vectors to obtain a set of bit-by-bit displacement entropies; Performing a weighting process based on a Softmax function on the set of bit-by-bit displacement entropies to obtain a set of bit-by-bit displacement entropy optimization factors; Based on the set of bit-by-bit displacement entropy optimization factors, the set of patient safety early warning judgment feature eigencomponent encoding vectors is fused to obtain an optimized patient safety early warning judgment feature vector.

9. A hospital patient safety management method based on data analysis, characterized in that: include: Obtaining the patient's hospitalization records, the patient's heart rate values collected by the smart device at multiple predetermined time points, and the patient's blood pressure values collected by the smart device at multiple predetermined time points; Extracting a patient hospitalization record-related feature vector and a patient state multimodal feature vector from the patient hospitalization record, the patient's heart rate values at multiple predetermined time points collected by the smart device, and the patient's blood pressure values at multiple predetermined time points collected by the smart device; Based on the patient hospitalization record associated feature vector and the patient status multimodal feature vector, it is determined whether to issue a patient safety warning.

10. The hospital patient safety management method based on data analysis according to claim 9, characterized in that: Extracting a patient hospitalization record-associated feature vector and a patient state multimodal feature vector from the patient hospitalization record, the patient's heart rate values at multiple predetermined time points collected by the smart device, and the patient's blood pressure values at multiple predetermined time points collected by the smart device, including: Performing feature extraction on the patient's hospitalization record to obtain a feature vector associated with the patient's hospitalization record; Performing feature extraction on the heart rate values of the patient at a plurality of predetermined time points collected by the smart device to obtain a heart rate feature vector of the patient; performing feature extraction on the blood pressure values of the patient at a plurality of predetermined time points collected by the smart device to obtain a blood pressure feature vector of the patient; The patient's heart rate feature vector and the patient's blood pressure feature vector are associated to obtain the patient's state multimodal feature vector.

Images