A gynecological clinical nursing safety hazard management system
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GENERAL HOSPITAL OF THE NORTHERN WAR ZONE OF THE CHINESE PEOPLES LIBERATION ARMY
- Filing Date
- 2026-02-04
- Publication Date
- 2026-05-29
AI Technical Summary
The existing obstetrics and gynecology nursing safety management system has shortcomings in data integration, risk identification, early warning design and handling procedures. It is difficult to meet the safety management needs of the whole scenario and the whole process. It has poor data compatibility with multiple systems, single risk assessment, lack of hierarchical and targeted early warning information and standardized handling procedures.
By employing multi-source data fusion technology, a gynecological and obstetric nursing safety hazard management system is constructed through heterogeneous system adaptation modules, hazard intelligent perception modules, risk quantification and assessment modules, hierarchical and precise early warning modules, and hazard hierarchical handling modules. This system enables format conversion, semantic alignment, and encrypted transmission of multi-source data, and combines a dedicated machine learning model for hazard identification. It also constructs a hierarchical and multi-dimensional risk assessment system and designs differentiated early warning rules and standardized handling procedures.
It has achieved comprehensiveness, precision and efficiency in nursing safety management, reduced the omission of potential hazards, shortened the risk response cycle, improved the ability to prevent and control risks in advance, and ensured the standardization and scientific nature of the hazard handling process.
Smart Images
Figure CN122117282A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent medical care technology, and in particular to a safety hazard management system for clinical nursing in obstetrics and gynecology. Background Technology
[0002] With the rapid iteration of medical informatization and intelligent technologies, the field of obstetrics and gynecology nursing safety management has gradually transformed from a traditional model to a technology-enabled one. Early management of nursing safety hazards relied on manual inspections, recording, and experience-based judgment by medical staff, which was inefficient and prone to omissions due to human error. The widespread adoption of hospital information systems and electronic medical record systems has promoted the electronic storage and initial integration of nursing data, providing basic data support for safety management. In recent years, technologies such as the Internet of Things and machine learning have begun to be deeply integrated into this field. Related systems are attempting to conduct risk warnings through data collection and algorithm analysis, promoting a shift in nursing safety management from passive response to proactive prevention. While the level of intelligence continues to improve, there is still room for optimization.
[0003] Existing technologies still have many shortcomings in practical applications, making it difficult to meet the complex and diverse safety management needs of obstetrics and gynecology nursing scenarios. The lack of unified data standards across different hospital systems hinders the efficient integration and semantic collaboration of multi-source heterogeneous data, preventing the full realization of data value. Risk identification often focuses on single symptoms or physiological indicators, failing to comprehensively cover diverse potential hazards such as environmental safety and equipment operation. Assessment indicators have limited dimensions and lack dynamic updating mechanisms. Early warning information pushes lack hierarchical and targeted design, easily leading to information overload or omission of key risks. Furthermore, there is a lack of standardized closed-loop procedures for hazard handling. For example, patent CN120531348A, while improving the accuracy of postpartum hemorrhage early warning by collecting blood flow data and vital sign information and using AI models to calculate bleeding risk levels and trigger hierarchical warnings, does not solve the problem of data compatibility across multiple systems. Risk assessment only focuses on the single symptom of postpartum hemorrhage, neglecting other high-frequency obstetric and gynecological hazards, and it lacks a standardized hazard handling process, failing to achieve comprehensive safety management across all scenarios and processes. Summary of the Invention
[0004] This invention aims to achieve intelligent management and control of safety hazards in obstetric and gynecological nursing throughout the entire process. By integrating multi-source data, accurately identifying hazards, conducting scientific risk assessments, providing differentiated early warnings, and standardizing procedures, it enhances the comprehensiveness, accuracy, and efficiency of nursing safety management.
[0005] The present invention adopts the following technical solution:
[0006] A safety hazard management system for obstetrics and gynecology clinical nursing includes a heterogeneous system adaptation module, a hazard intelligent perception module, a risk quantification and assessment module, a graded and precise early warning module, a hazard graded handling module, and a data intelligent analysis module. It performs intelligent management and control of safety hazards throughout the entire process of obstetrics and gynecology nursing. The heterogeneous system adaptation module interfaces with hospital information systems, electronic medical record systems, maternal and infant monitoring systems, and disinfection supply systems, performing format conversion, semantic alignment, and encrypted transmission of multi-source heterogeneous data. The hazard intelligent perception module collects structured business data and unstructured environmental and equipment data, integrates them, and analyzes and identifies potential safety hazards using a dedicated machine learning model to trigger alarms. A combination of random forest and gradient boosting tree algorithms is used to adapt to the construction of obstetrics and gynecology nursing scenarios; the risk quantification assessment module constructs a hierarchical multi-dimensional indicator system, completes the determination of the risk level of hidden dangers based on the quantitative analysis results of the indicators, and sets up a dynamic update mechanism for the risk level; the graded and precise early warning module configures differentiated early warning rules according to the risk level, and performs targeted push of early warning information, full-process traceability, and timeout escalation operations; the hidden danger graded disposal module embeds standardized disposal processes, performs hidden danger disposal task dispatch, real-time progress tracking, and disposal effect verification and archiving; the data intelligent analysis module adopts a hybrid storage architecture to store full-process control data, and uses algorithms to mine the occurrence patterns of hidden dangers and output visualized analysis results.
[0007] Furthermore, the dedicated machine learning model in the hazard intelligent perception module employs a combination of random forest and gradient boosting tree algorithms, sets up a local feature extraction layer, and designs separate feature matching branches for two high-frequency hazards: neonatal identification errors and postpartum infections. Model training samples are categorized and labeled according to prenatal, intrapartum, and postpartum clinical pathways. The proportion of dedicated samples is adapted to the model training accuracy, and the proportion of dedicated samples is higher than that of general medical samples. Model parameters are optimized through k-fold cross-validation, with the number of folds in the k-fold cross-validation adapted to the model training sample size. A periodic iterative training mechanism is also implemented, and the hazard identification probability is calculated using the following formula:
[0008] ;
[0009] in, To determine the probability of identifying potential hazards, For structured data feature matching coefficients, For the first Item structured data feature values, To adapt weights to unstructured data samples, For the first Unstructured data feature values, For the number of features in structured data, The data consists of unstructured data features. Structured data originates from business systems connected to heterogeneous system adaptation modules, while unstructured data originates from environmental and equipment data collected by the hazard intelligent sensing module.
[0010] Furthermore, the risk quantification assessment module constructs a hierarchical, multi-dimensional risk assessment indicator system. This system uses primary indicators as its core framework, with secondary indicators serving as detailed extensions of the primary indicators. Primary assessment indicators include the likelihood of a potential hazard occurring, the degree of impact on patient safety, the degree of interference with nursing care, and the rate of hazard spread. Each primary indicator is further subdivided into several secondary indicators. The weight of each indicator is determined using the analytic hierarchy process (AHP), and the overall risk score is calculated using the following formula:
[0011] ;
[0012] in, To calculate the overall risk score, For the first The weight values of each evaluation indicator, For the first The quantitative score of each evaluation indicator, To assess the total number of indicators,
[0013] Risk levels are determined based on comprehensive risk scores. The triggering conditions for level updates include two scenarios: changes in hazard-related data and updates to the handling progress. Once a level changes, a notification is sent to the corresponding management personnel, and the notification is distributed through the hierarchical and precise early warning module.
[0014] Furthermore, the differentiated early warning rules of the tiered and precise early warning module are set in layers according to risk levels, which are divided into general risk, major risk, serious risk, and extremely serious risk. General risks are only pushed to the responsible nurse through system pop-ups; major risks are pushed to the head nurse and responsible nurses simultaneously through system pop-ups and voice broadcasts; serious risks are pushed to the department director, head nurse, and responsible nurses simultaneously through system pop-ups, voice broadcasts, and mobile application messages; extremely serious risks also push early warning information to the director of the nursing department via SMS, and simultaneously activate the hospital-wide emergency response process. The early warning information records the generation time, push recipients, reception status, and start and end times of handling. If no handling is carried out within the time limit, the early warning level is upgraded. The early warning level upgrade rules correspond to and are adapted to the differentiated early warning rules.
[0015] Furthermore, the heterogeneous system adaptation module supports HL7, DICOM, and RESTful protocols. It performs interface registration, authentication, monitoring, and rate limiting through the interface gateway, uses SSL / TLS encryption technology to handle data transmission, and has a built-in data format conversion engine and data semantic mapping library. It uses natural language processing technology to perform semantic parsing on unstructured nursing text data, extracts key information, and converts it into structured data. It performs real-time monitoring of the interface's operating status and triggers maintenance notifications when the interface is abnormal. The interface gateway establishes stable connections with the hospital information system, electronic medical record system, maternal and infant monitoring system, and sterilization supply system. The data semantic mapping library performs unified mapping for the same business terms in different systems.
[0016] Furthermore, the hazard intelligent sensing module deploys an IoT sensor network, which includes temperature and humidity sensors, ground slippage sensors, and equipment operation status sensors. The temperature and humidity sensors are configured with different collection frequencies according to the functional type of the deployment scenario. The temperature and humidity sensors deployed in wards and treatment rooms have different collection frequencies. The unstructured data collected through the IoT sensor network is preprocessed and then fused with the structured data transmitted by the heterogeneous system adaptation module. The preprocessing uses a moving average noise reduction and threshold calibration algorithm. The fused data is then input into a dedicated machine learning model for analysis. The threshold calibration algorithm is adapted to the feature requirements of the dedicated machine learning model.
[0017] Furthermore, the adaptive learning rate of the dedicated machine learning model is calculated using the following formula:
[0018] ;
[0019] in, For adaptive learning rate, The initial learning rate, The number of training iterations for the model. As a decay coefficient, the weight of the newborn's age feature is set as a fixed coefficient for the newborn fall risk identification scenario.
[0020] Furthermore, the standardized process of the hazard classification and handling module is embedded in the system in the form of a visual flowchart, clearly defining the operation steps, precautions, and required material list. It also includes a pre-set list of job responsibilities, assigning handling tasks according to hazard type and risk level, with real-time updates of task progress. After handling is completed, evidence is uploaded, and the system automatically verifies it according to pre-set verification standards. Once verification is successful, the entire process data is archived to the data intelligent analysis module. The data intelligent analysis module adopts a hybrid storage architecture combining relational and non-relational databases, encrypts sensitive data, sets three levels of user permissions, establishes a regular data backup and disaster recovery mechanism, and uses association rule mining algorithms and time series analysis algorithms to uncover the patterns and trends of hazard occurrence. Sensitive data includes maternal privacy and newborn information, and the three levels of user permissions correspond to administrator, head nurse, and responsible nurse, respectively.
[0021] The advantages of this invention are:
[0022] This invention integrates multi-system data transmission and processing technologies with a dedicated machine learning model adapted to obstetrics and gynecology scenarios, and combines various types of data to conduct hazard analysis. This enables accurate identification and rapid alarm of potential nursing hazards, reduces hazard omissions, makes hazard discovery more timely and comprehensive, and improves the proactive prevention and control capabilities of nursing safety management.
[0023] This invention constructs a multi-dimensional risk assessment system, scientifically classifies risk levels and establishes a dynamic update mechanism, and combines it with differentiated early warning push and timeout upgrade functions to achieve reasonable determination of risk levels and accurate delivery of early warning information. Medical staff at different levels can quickly respond to corresponding risks, significantly shorten the risk response cycle, and enhance the pertinence of risk management.
[0024] This invention embeds a visual standardized handling process, accurately assigns handling tasks and tracks and verifies them in real time. By combining hybrid storage and data mining technology to uncover patterns in potential hazards, it achieves standardized and efficient operation of the hazard handling process, effectively reduces the probability of hazard recurrence, and makes nursing safety management decisions more scientific and targeted. Attached Figure Description
[0025] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application.
[0026] In the attached diagram:
[0027] Figure 1 This is a system framework diagram of a safety hazard management system for obstetric and gynecological clinical nursing in Example 1. Detailed Implementation
[0028] The present invention will now be described in detail and specifically through specific embodiments to enable a better understanding of the invention. However, the following embodiments do not limit the scope of protection of the present invention.
[0029] Example 1
[0030] like Figure 1 As shown, a safety hazard management system for obstetrics and gynecology clinical nursing includes a heterogeneous system adaptation module, a hazard intelligent perception module, a risk quantification and assessment module, a graded and precise early warning module, a hazard graded handling module, and a data intelligent analysis module. This system enables intelligent management and control of safety hazards throughout the entire obstetrics and gynecology nursing process. The heterogeneous system adaptation module interfaces with hospital information systems, electronic medical record systems, maternal and infant monitoring systems, and disinfection supply systems, performing format conversion, semantic alignment, and encrypted transmission of multi-source heterogeneous data. The hazard intelligent perception module collects structured business data and unstructured environmental and equipment data, integrates them, and analyzes and identifies potential safety hazards using a dedicated machine learning model, triggering alarms. The system employs a combination of random forest and gradient boosting tree algorithms, adapted for obstetrics and gynecology nursing scenarios. The risk quantification assessment module constructs a hierarchical, multi-dimensional indicator system, determines the risk level of potential hazards based on the quantitative analysis results, and sets up a dynamic risk level update mechanism. The graded and precise early warning module configures differentiated early warning rules according to the risk level, executes targeted push notifications of early warning information, full-process traceability, and timeout escalation operations. The hazard graded handling module embeds standardized handling procedures, executes hazard handling task assignment, real-time progress tracking, and verification and archiving of handling effects. The data intelligent analysis module uses a hybrid storage architecture to store full-process control data, mines the patterns of hazard occurrence through algorithms, and outputs visualized analysis results.
[0031] In a specific embodiment, the entire process of postpartum care for mothers in the obstetrics and gynecology ward of a hospital, from admission to postpartum recovery and discharge, is taken as the complete application scenario. The system's six major modules work together to construct a closed-loop management and control link. The heterogeneous system adaptation module connects in advance with the hospital's information system for mothers' identity information, the electronic medical record system for prenatal check-up records, the maternal and infant monitoring system for fetal heart rate monitoring data, and the sterilization information of hospital bed instruments from the sterilization supply system. Addressing the differences in data formats across systems, the built-in conversion engine unifies formats such as XML and JSON into a standard format. Semantic mapping technology standardizes terminology across different systems, and encrypted transmission technology is used throughout the process to ensure mothers' privacy and the security of medical data.
[0032] The hazard intelligent perception module continuously collects structured data such as body temperature, blood pressure, and delivery records, as well as unstructured data such as ward temperature and humidity, floor conditions, and fetal heart monitor operating parameters during the mother's hospitalization. After the two types of data are fused, they are input into a combination model of random forest and gradient boosting tree optimized for obstetrics and gynecology scenarios. This model takes into account both feature extraction capability and classification accuracy, and can keenly capture potential hazards and trigger alarms.
[0033] The risk quantification assessment module establishes a hierarchical, multi-dimensional indicator system, focusing on the likelihood of hazard occurrence, the degree of impact, the level of interference, and the speed of spread. A dynamic update mechanism adjusts the level according to changes in hazard data and the progress of handling, avoiding the lag of static assessments. The graded, precise early warning module customizes push rules based on risk levels, delivering information in a targeted manner and recording the entire process trajectory. An upgrade design for exceeding timeout periods ensures that risks are not overlooked.
[0034] The hazard classification and handling module presents standardized processes intuitively, clearly defining task allocation and progress tracking nodes. Post-handling verification and archiving form a closed-loop management system. The data intelligent analysis module's hybrid storage architecture adapts to different types of data storage needs. Hazard patterns discovered by algorithms are presented in a visual format, providing an intuitive reference for optimizing nursing safety management throughout the entire maternity hospitalization cycle.
[0035] Furthermore, the dedicated machine learning model in the hazard intelligent perception module employs a combination of random forest and gradient boosting tree algorithms, sets up a local feature extraction layer, and designs separate feature matching branches for two high-frequency hazards: neonatal identification errors and postpartum infections. Model training samples are categorized and labeled according to prenatal, intrapartum, and postpartum clinical pathways. The proportion of dedicated samples is adapted to the model training accuracy, and the proportion of dedicated samples is higher than that of general medical samples. Model parameters are optimized through k-fold cross-validation, with the number of folds in the k-fold cross-validation adapted to the model training sample size. A periodic iterative training mechanism is also implemented, and the hazard identification probability is calculated using the following formula:
[0036] ;
[0037] in, To determine the probability of identifying potential hazards, For structured data feature matching coefficients, For the first Item structured data feature values, To adapt weights to unstructured data samples, For the first Unstructured data feature values, For the number of features in structured data, The data consists of unstructured data features. Structured data originates from business systems connected to heterogeneous system adaptation modules, while unstructured data originates from environmental and equipment data collected by the hazard intelligent sensing module.
[0038] In specific embodiments, within the aforementioned full-cycle maternal care scenario, the combined algorithm of the dedicated machine learning model plays a crucial role. Random forests effectively handle high-dimensional data and reduce the risk of overfitting, while gradient boosting trees excel at gradually correcting prediction biases. The combination of these two approaches allows the model to maintain stable performance in complex data scenarios. A local feature extraction layer specifically captures key features closely related to potential risks. Separate feature matching branches are designed for two common obstetric risks: neonatal identification errors and puerperal infections, enhancing the identification capability of core risks. Model training samples are categorized and labeled according to prenatal, intrapartum, and postpartum clinical pathways. Prenatal samples include 2000 data points related to pregnancy complications, intrapartum samples focus on indicators of the delivery process (1500 data points), and postpartum samples emphasize recovery information (1500 data points). The proportion of dedicated samples is higher than that of general medical samples, allowing the model to learn more deeply the specific patterns of obstetric and gynecological scenarios. k-fold cross-validation adjusts the number of folds based on the total number of samples, optimizing model parameters through multiple rounds of validation. A regular iterative training mechanism incorporates new clinical data monthly to adapt to subtle changes in the nursing scenario. In the probability formula for hazard identification, the values of and are determined based on the actual contribution of two types of data to hazard identification. The structured information, such as maternal gestational age, fetal heart rate monitoring data, and blood routine indicators, transmitted by the heterogeneous system, corresponds to the unstructured data, such as ward temperature and humidity, ventilation frequency, and equipment operating parameters collected by the sensor. The formula sets the number of structured features to 25 and the number of unstructured features to 12. This formula transforms multi-dimensional data into intuitive probability values, and calculates the values of various hazards in real time during the maternal hospitalization.
[0039] Furthermore, the risk quantification assessment module constructs a hierarchical, multi-dimensional risk assessment indicator system. This system uses primary indicators as its core framework, with secondary indicators serving as detailed extensions of the primary indicators. Primary assessment indicators include the likelihood of a potential hazard occurring, the degree of impact on patient safety, the degree of interference with nursing care, and the rate of hazard spread. Each primary indicator is further subdivided into several secondary indicators. The weight of each indicator is determined using the analytic hierarchy process (AHP), and the overall risk score is calculated using the following formula:
[0040] ;
[0041] in, To calculate the overall risk score, For the first The weight values of each evaluation indicator, For the first The quantitative score of each evaluation indicator, To assess the total number of indicators,
[0042] Risk levels are determined based on comprehensive risk scores. The triggering conditions for level updates include two scenarios: changes in hazard-related data and updates to the handling progress. Once a level changes, a notification is sent to the corresponding management personnel, and the notification is distributed through the hierarchical and precise early warning module.
[0043] In a specific embodiment, within the context of full-cycle maternal care, the hierarchical, multi-dimensional indicator system of the risk quantification assessment module is constructed around the core concerns of obstetric and gynecological nursing safety. The primary indicators focus on four key dimensions: the likelihood of a potential hazard occurring, its impact, the degree of interference, and the speed of its spread. Each primary indicator is further subdivided into three secondary indicators. For example, the likelihood of a potential hazard includes historical frequency of occurrence, trends in related factors, and the completeness of protective measures, making risk assessment more operational. The application of the analytic hierarchy process (AHP) incorporates the practical experience of obstetric and gynecological nursing experts, highlighting the impact of key indicators through scientific weight allocation. The weight of the impact on patient safety is higher than other indicators, making the assessment results more aligned with clinical reality. The comprehensive risk score formula reflects the differences in the importance of each indicator, representing a score quantified according to a unified standard. For example, 1 point corresponds to never occurring, and 3 points correspond to frequently occurring. The total number of 12 assessment indicators is set, and this formula transforms the multi-dimensional assessment content into a unified score, providing an objective basis for risk level classification. Risk levels are categorized based on the range of comprehensive scores. During a mother's hospitalization, when her body temperature data is updated every 10 minutes or when the handling of a potential hazard changes from in progress to completion, the system automatically recalculates the risk level (S) and adjusts the level accordingly. The updated level is promptly pushed to the head nurse through the tiered and precise early warning module, allowing management personnel to monitor risk dynamics in real time and providing timely support for decision-making in maternal care.
[0044] Furthermore, the differentiated early warning rules of the tiered and precise early warning module are set in layers according to risk levels, which are divided into general risk, major risk, serious risk, and extremely serious risk. General risks are only pushed to the responsible nurse through system pop-ups; major risks are pushed to the head nurse and responsible nurses simultaneously through system pop-ups and voice broadcasts; serious risks are pushed to the department director, head nurse, and responsible nurses simultaneously through system pop-ups, voice broadcasts, and mobile application messages; extremely serious risks also push early warning information to the director of the nursing department via SMS, and simultaneously activate the hospital-wide emergency response process. The early warning information records the generation time, push recipients, reception status, and start and end times of handling. If no handling is carried out within the time limit, the early warning level is upgraded. The early warning level upgrade rules correspond to and are adapted to the differentiated early warning rules.
[0045] In a specific embodiment, within the complete scenario of full-cycle maternal care, the differentiated rules of the graded and precise early warning module are designed based on the impact range of the risk level and the priority of treatment. When the temperature and humidity in the ward slightly exceed the standard range and are determined to be of general risk, the system only pushes a pop-up warning to the responsible nurse, prompting them to adjust the air conditioning parameters in a timely manner. When the parameters of the neonatal fetal heart rate monitoring equipment show a small fluctuation and are determined to be of greater risk, the system simultaneously notifies the head nurse and the responsible nurse through pop-ups and voice broadcasts, urging them to check the equipment status. When a postpartum woman experiences unexplained bleeding precursors and is determined to be of major risk, the system notifies the department director, head nurse, and responsible nurse through three channels: pop-ups, voice broadcasts, and mobile application messages, ensuring a rapid response. When a suspected cluster of infections occurs in the ward and is determined to be of particularly serious risk, the system adds SMS notification to the director of the nursing department on top of the above channels, simultaneously activating the hospital-wide emergency response process and allocating relevant resources such as the infection control department. All early warning information is recorded in detail, including the generation time, the recipient, the receiving status, and the start and end times of the handling. Different risk levels are assigned corresponding handling time limits: 30 minutes for general risks and 20 minutes for major risks. If no action is taken within the time limit, the early warning level will be automatically upgraded and the scope of notification expanded, ensuring that all types of risks can be responded to and handled in a timely manner throughout the entire process of a pregnant woman's hospitalization.
[0046] Furthermore, the heterogeneous system adaptation module supports HL7, DICOM, and RESTful protocols. It performs interface registration, authentication, monitoring, and rate limiting through the interface gateway, uses SSL / TLS encryption technology to handle data transmission, and has a built-in data format conversion engine and data semantic mapping library. It uses natural language processing technology to perform semantic parsing on unstructured nursing text data, extracts key information, and converts it into structured data. It performs real-time monitoring of the interface's operating status and triggers maintenance notifications when the interface is abnormal. The interface gateway establishes stable connections with the hospital information system, electronic medical record system, maternal and infant monitoring system, and sterilization supply system. The data semantic mapping library performs unified mapping for the same business terms in different systems.
[0047] In a specific embodiment, throughout the entire process from admission to discharge, the heterogeneous system adaptation module supports multiple mainstream medical-related protocols. The HL7 protocol adapts to text-based medical data transmission, the DICOM protocol meets the needs of neonatal ultrasound image data interaction, and the RESTful protocol adapts to various API calls, ensuring smooth integration with different hospital business systems. The interface gateway handles interface registration, authentication, monitoring, and rate limiting. Registration and authentication ensure secure interface access, monitoring tracks interface operation status every 10 seconds, and rate limiting sets a maximum of 50 concurrent requests per interface to prevent data transmission overload and system anomalies. SSL / TLS encryption is applied throughout data transmission, effectively preventing the leakage or tampering of sensitive information such as maternal pregnancy records and neonatal health data. The data format conversion engine automatically identifies data formats from different systems and converts them to a unified standard format. The data semantic mapping library includes over 300 commonly used obstetrics and gynecology business terms, providing a unified mapping for terms with different expressions but consistent meanings across different systems, eliminating semantic ambiguity. Natural language processing technology extracts key information such as the mother's diet and activity level from unstructured texts such as nurses' handwritten postpartum care logs through semantic parsing, converts them into structured data and integrates them into the overall data system. This provides complete data support for subsequent hazard identification and risk assessment throughout the mother's hospital stay. When the interface of the sterilization supply system is interrupted, the system automatically attempts to reconnect. After 10 failed attempts, a maintenance notification is triggered to ensure stable data interaction.
[0048] Furthermore, the hazard intelligent sensing module deploys an IoT sensor network, which includes temperature and humidity sensors, ground slippage sensors, and equipment operation status sensors. The temperature and humidity sensors are configured with different collection frequencies according to the functional type of the deployment scenario. The temperature and humidity sensors deployed in wards and treatment rooms have different collection frequencies. The unstructured data collected through the IoT sensor network is preprocessed and then fused with the structured data transmitted by the heterogeneous system adaptation module. The preprocessing uses a moving average noise reduction and threshold calibration algorithm. The fused data is then input into a dedicated machine learning model for analysis. The threshold calibration algorithm is adapted to the feature requirements of the dedicated machine learning model.
[0049] In a specific embodiment, in the whole-cycle postpartum care scenario, the IoT sensor network of the potential hazard intelligent perception module is selected and deployed according to the scenario requirements. High-precision temperature and humidity sensors are selected, with one sensor deployed next to every two beds in the ward, and the data collection frequency set to once per minute, focusing on monitoring the postpartum woman's resting environment. In the treatment room, where aseptic procedures are involved, two temperature and humidity sensors are deployed, with the data collection frequency increased to once every 30 seconds. Infrared sensors are used for floor slippage, and are primarily deployed in areas prone to water accumulation, such as delivery rooms, bathroom entrances, and corridor corners, to monitor floor conditions in real time and prevent falls. Equipment operation status sensors are directly connected to key medical equipment such as fetal heart monitors, infusion pumps, and disinfection equipment, collecting parameters such as operating voltage, operating temperature, and operating noise to promptly detect potential equipment malfunctions. During the mother's hospitalization, the sensor network continuously collects unstructured data. In the preprocessing stage, a moving average denoising algorithm uses a sliding window of size 5 to filter out random errors during sensor acquisition. A threshold calibration algorithm sets reasonable thresholds based on obstetric and gynecological nursing standards, such as ward temperature and humidity thresholds of 22-26 degrees Celsius and 40%-60% humidity, ensuring the data meets the characteristic requirements of the model analysis. The preprocessed unstructured data is then fused with structured data such as the mother's gestational age and prenatal examination results transmitted from heterogeneous systems to form a 37-dimensional data matrix. This matrix is continuously input into a dedicated machine learning model throughout the mother's hospitalization, enabling more comprehensive and accurate identification of potential risks and improving the detection efficiency of the sensing module.
[0050] Furthermore, the adaptive learning rate of the dedicated machine learning model is calculated using the following formula:
[0051] ;
[0052] in, For adaptive learning rate, The initial learning rate, The number of training iterations for the model. As a decay coefficient, the weight of the newborn's age feature is set as a fixed coefficient for the newborn fall risk identification scenario.
[0053] In a specific embodiment, in the context of full-cycle maternal care, the adaptive learning rate formula of the dedicated machine learning model is designed to align with the model training process. Initially, a larger learning rate is needed to quickly converge and approach the optimal solution. As the number of iterations increases, the learning rate needs to be gradually reduced to optimize parameter accuracy; therefore, this inverse proportional decay formula is used for dynamic adjustment. The formula uses a value of 0.1, combined with the initial model state setting, representing the number of training iterations (range 1-200), and λ of 0.001 to control the learning rate decay rate. The combination of these three factors ensures the model maintains good learning performance at different training stages. For the scenario of newborn fall risk identification, during the postpartum neonatal hospitalization period, the newborn's age directly correlates with activity level and fall risk. Younger newborns have less independent activity and lower risk, while older newborns have increased activity and higher risk. Therefore, the weight of this feature is set to a fixed coefficient of 0.3, allowing the model to focus on this key factor when analyzing such potential risks. When a 6-month-old newborn tries to roll over in the ward, the activity data collected by the sensors is fused with the structured data such as the newborn's age and weight transmitted by the heterogeneous system and then input into the model. The model, optimized by the adaptive learning rate and combined with fixed age feature weights, quickly and accurately identifies the risk of falling and triggers an alarm in a timely manner, thus providing protection for the newborn's safety.
[0054] Furthermore, the standardized process of the hazard classification and handling module is embedded in the system in the form of a visual flowchart, clearly defining the operation steps, precautions, and required material list. It also includes a pre-set list of job responsibilities, assigning handling tasks according to hazard type and risk level, with real-time updates of task progress. After handling is completed, evidence is uploaded, and the system automatically verifies it according to pre-set verification standards. Once verification is successful, the entire process data is archived to the data intelligent analysis module. The data intelligent analysis module adopts a hybrid storage architecture combining relational and non-relational databases, encrypts sensitive data, sets three levels of user permissions, establishes a regular data backup and disaster recovery mechanism, and uses association rule mining algorithms and time series analysis algorithms to uncover the patterns and trends of hazard occurrence. Sensitive data includes maternal privacy and newborn information, and the three levels of user permissions correspond to administrator, head nurse, and responsible nurse, respectively.
[0055] In a specific embodiment, within the complete scenario of full-cycle maternal care, the standardized process of the hazard classification and handling module is presented through an intuitive visual flowchart, clearly defining the operational steps according to the logical sequence of hazard handling. For example, when a slippery floor is detected in the maternal ward, the responsible nurse, upon receiving an alert, follows the steps shown in the flowchart, carrying anti-slip warning signs, mops, absorbent towels, and other supplies to the site. First, a warning sign is placed 1 meter in front of the hazard area. Then, the absorbent towel is used to absorb the water, and the floor is wiped dry with the mop, strictly adhering to precautions to prevent the spread of water stains. A pre-set list of job responsibilities clearly defines the duties of the administrator, head nurse, and responsible nurse. The administrator coordinates, the head nurse supervises and guides, and the responsible nurse executes on-site. The system accurately assigns tasks according to the hazard type and risk level, and task progress is synchronized in real time, allowing administrators to view progress at any time. After handling, the responsible nurse takes a photo of the dry area and uploads it to the system. The system automatically verifies according to preset standards, completing the verification after confirming there is no standing water or slippery feeling. All process data is completely archived to the data intelligent analysis module. In this module's hybrid storage architecture, a relational database stores structured data such as basic maternal information and risk assessment scores, while a non-relational database stores unstructured data such as raw sensor data and photos of the treatment site. Sensitive data is stored using AES-256 encryption technology, and access is strictly divided into three levels of user permissions: administrators can view all data and configure the system, head nurses can view relevant data for their department, and responsible nurses can only access data for the patients they are responsible for. Regular data backups are performed weekly, and disaster recovery employs an off-site backup solution to ensure data security. Association rule mining algorithms have discovered the intrinsic link between ward temperature and humidity and postpartum infections, while time series analysis algorithms have captured the high-incidence period of neonatal jaundice risk 3-7 days postpartum. The mining results are presented in visual charts, providing a powerful reference for the department to adjust nursing plans throughout the entire maternal hospitalization period.
[0056] The specific embodiments of the present invention have been described in detail above, but they are merely examples, and the present invention is not equivalent to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions to the present invention are also within the scope of the present invention. Therefore, all equivalent transformations and modifications made without departing from the spirit and scope of the present invention should be covered within the scope of the present invention.
Claims
1. A safety hazard management system for obstetrics and gynecology clinical nursing, characterized in that, The system includes a heterogeneous system adaptation module, a hidden danger intelligent perception module, a risk quantification and assessment module, a graded and precise early warning module, a hidden danger graded handling module, and a data intelligent analysis module, which implements intelligent management and control of safety hazards throughout the entire process of obstetrics and gynecology nursing. The heterogeneous system adaptation module interfaces with hospital information systems, electronic medical record systems, maternal and infant monitoring systems, and disinfection supply systems, and performs format conversion, semantic alignment, and encrypted transmission of multi-source heterogeneous data. The hidden danger intelligent perception module collects structured business data and unstructured environmental and equipment data, and after fusion, analyzes and identifies potential safety hazards and triggers alarms through a dedicated machine learning model. The dedicated machine learning model adopts a combination algorithm of random forest and gradient boosting tree, adapted to the obstetrics and gynecology nursing scenario. The risk quantification assessment module constructs a hierarchical, multi-dimensional indicator system, determines the risk level of potential hazards based on the quantitative analysis results of the indicators, and sets up a dynamic risk level update mechanism. The graded and precise early warning module configures differentiated early warning rules according to the risk level, and performs targeted push of early warning information, full-process traceability, and timeout escalation operations. The hazard graded disposal module embeds standardized disposal processes, performs hazard disposal task dispatching, real-time progress tracking, and disposal effect verification and archiving. The data intelligent analysis module uses a hybrid storage architecture to store full-process control data, mines the patterns of hazard occurrence through algorithms, and outputs visualized analysis results.
2. The obstetric and gynecological clinical nursing safety hazard management system according to claim 1, wherein the dedicated machine learning model in the hazard intelligent perception module adopts a combination algorithm of random forest and gradient boosting tree, sets a local feature extraction layer, and designs separate feature matching branches for two high-frequency hazards: newborn identification errors and puerperal infections. The model training samples are classified and labeled according to prenatal, intrapartum, and postpartum clinical pathways. The proportion of the dedicated samples is adapted to the model training accuracy, and the proportion of the dedicated samples is higher than the proportion of general medical samples. The model parameters are optimized through k-fold cross-validation, and the number of folds in the k-fold cross-validation is adapted to the model training sample size. At the same time, a periodic iterative training mechanism is set, and the hazard identification probability is calculated according to the formula: ； in, To determine the probability of identifying potential hazards, For structured data feature matching coefficients, For the first Item structured data feature values, To adapt weights to unstructured data samples, For the first Unstructured data feature values, For the number of features in structured data, The number of unstructured data features refers to the structured data originating from the business system connected to the heterogeneous system adaptation module, while the unstructured data originates from the environmental and equipment data collected by the hidden danger intelligent perception module.
3. The obstetrics and gynecology clinical nursing safety hazard management system according to claim 2, wherein the risk quantification assessment module constructs a hierarchical multi-dimensional risk assessment indicator system, wherein the hierarchical multi-dimensional risk assessment indicator system takes primary indicators as the core framework, and secondary indicators are detailed extensions of primary indicators. The primary assessment indicators include the probability of hazard occurrence, the degree of impact on patient safety, the degree of interference with nursing work, and the speed of hazard spread. Each primary indicator is further subdivided into several secondary indicators. The weight of each indicator is determined by the analytic hierarchy process (AHP), and the comprehensive risk score is calculated according to the formula: ； in, To calculate the overall risk score, For the first The weight values of each evaluation indicator, For the first The quantitative score of each evaluation indicator, To assess the total number of indicators, Risk levels are classified according to comprehensive risk scores. The triggering conditions for level updates include two scenarios: changes in hazard-related data and updates to the handling progress. After a level change, a notification is pushed to the corresponding management personnel, and the notification is issued through the hierarchical and precise early warning module.
4. The obstetrics and gynecology clinical nursing safety hazard management system according to claim 3, wherein the differentiated early warning rules of the graded precision early warning module are set according to risk level, and the risk level is divided into general risk, major risk, serious risk, and particularly serious risk; The general risks are only alerted to the responsible nurse via a system pop-up window. The significant risks are alerted to the head nurse and the nurse in charge through a system pop-up and voice broadcast simultaneously; the major risks are alerted to the department director, head nurse and nurse in charge simultaneously through a system pop-up, voice broadcast and mobile application message simultaneously. The aforementioned particularly serious risks will trigger an SMS warning to the head of the nursing department, simultaneously initiating the hospital-wide emergency response process. The warning information will record the generation time, recipient, reception status, and start and end times of the response. If no response is taken within the specified time, the warning level will be upgraded. The rules for upgrading the warning level will correspond to and be adapted to the differentiated warning rules.
5. A safety hazard management system for obstetric and gynecological clinical nursing according to claim 4, wherein the heterogeneous system adaptation module supports HL7, DICOM and RESTful protocols, performs interface registration, authentication, monitoring and rate limiting through the interface gateway, uses SSL / TLS encryption technology to process data transmission, has a built-in data format conversion engine and data semantic mapping library, performs semantic parsing of unstructured nursing text data through natural language processing technology, extracts key information and converts it into structured data, performs real-time monitoring of interface operation status, triggers operation and maintenance notification when the interface is abnormal, the interface gateway establishes stable connections with hospital information system, electronic medical record system, maternal and infant monitoring system and disinfection supply system, and the data semantic mapping library performs unified mapping for the same business terms in different systems.
6. A safety hazard management system for obstetric and gynecological clinical nursing according to claim 5, wherein the hazard intelligent sensing module deploys an Internet of Things (IoT) sensor network, the IoT sensor network includes temperature and humidity sensors, ground slippage sensors, and equipment operation status sensors, the temperature and humidity sensors are set with a collection frequency according to the functional type of the deployment scenario, the temperature and humidity sensors deployed in wards and treatment rooms have different collection frequencies, the unstructured data collected through the IoT sensor network is preprocessed and then fused with the structured data transmitted by the heterogeneous system adaptation module, the preprocessing adopts a moving average denoising and threshold calibration algorithm, the fused data is input into a dedicated machine learning model for analysis, and the threshold calibration algorithm is adapted to the feature requirements of the dedicated machine learning model.
7. In the obstetrics and gynecology clinical nursing safety hazard management system according to claim 6, the adaptive learning rate of the dedicated machine learning model is calculated according to the formula: ； in, For adaptive learning rate, The initial learning rate, The number of training iterations for the model. As a decay coefficient, the weight of the newborn's age feature is set as a fixed coefficient for the newborn fall risk identification scenario.
8. A safety hazard management system for obstetric and gynecological clinical nursing according to claim 7, wherein the standardized process of the hazard classification and disposal module is embedded in the system in the form of a visual flowchart, clearly defining the operation steps, precautions, and required material list, pre-setting a job responsibility list, assigning disposal tasks according to the hazard type and risk level, the task progress is updated synchronously in real time, disposal evidence is uploaded after disposal is completed, the system automatically verifies according to preset verification standards, and after verification, the entire process data is archived to the data intelligent analysis module; the data intelligent analysis module adopts a hybrid storage architecture combining relational databases and non-relational databases, performs encrypted storage on sensitive data, sets three levels of user permissions, establishes a regular data backup and disaster recovery mechanism, and mines the occurrence patterns and trends of hazards through association rule mining algorithms and time series analysis algorithms, wherein the sensitive data includes maternal privacy and newborn information, and the three levels of user permissions correspond to the administrator, the head nurse, and the responsible nurse, respectively.