Internet-of-things-based monitoring ward grading nursing digital management method

Abstract

The invention relates to the technical field of biomedicine, in particular to an internet-of-things-based digitized management method for graded nursing of an intensive care unit. The method comprises the following steps that heterogeneous data of a patient are collected in real time through nursing monitoring equipment, and a principle physiological data flow is obtained; performing comprehensive physiological index calculation on the principle physiological data flow to obtain a comprehensive physiological index; carrying out individualized physiological index prediction according to the comprehensive physiological index, and carrying out patient condition deterioration probability prediction to obtain condition deterioration probability data; performing nursing level demand prediction according to the individualized physiological index prediction result to obtain nursing level demand probability data; and performing illness state trend prediction vector generation on the illness state deterioration probability data and the nursing level demand probability data to obtain an illness state trend prediction vector. According to the invention, through more accurate illness state trend prediction and graded early warning, medical workers are better assisted to carry out graded nursing on patients, and the management level of an intensive care unit is improved.

Description

Technical Field

[0001] The present invention relates to the field of biomedical technology, and in particular to a digital management method for graded nursing in an intensive care unit based on the Internet of Things. Background Art

[0002] As the most intensive department in the hospital for patient monitoring and care, the ICU has an urgent need for real-time, accurate and personalized care. The traditional ICU management model mainly relies on manual records and observations, which are inefficient, error-prone, and difficult to meet the growing medical needs. Therefore, the digital management method of graded care in ICUs based on the Internet of Things came into being. The popularity of various physiological parameter sensors, wearable devices and bedside nursing monitoring equipment makes it possible to collect multi-dimensional data of patients in real time. The development of computing, big data and artificial intelligence technologies makes it possible to process and analyze massive patient data in real time, providing a technical basis for disease prediction and early warning.

[0003] However, existing methods still have some shortcomings, such as inaccurate prediction of disease trends and inaccurate graded warnings. Some methods only consider a single or a few physiological indicators, while ignoring other factors that may affect the development of the disease, such as the patient's age, medical history, medication, etc. The lack of integration of multi-dimensional data leads to insufficient information in the prediction model, affecting the accuracy of the prediction. Some methods only warn of a single indicator, while ignoring the correlation between indicators. For example, a patient's heart rate and blood pressure are usually correlated with each other. If only one of the indicators is considered, some important warning signals may be missed. Summary of the invention

[0004] Based on this, it is necessary to provide a digital management method for graded nursing in intensive care units based on the Internet of Things to solve at least one of the above technical problems.

[0005] To achieve the above purpose, a digital management method for graded nursing care in intensive care units based on the Internet of Things includes the following steps:

[0006] Step S1: real-time collection of heterogeneous data of patients by nursing monitoring equipment to obtain principle physiological data streams; comprehensive physiological index calculations are performed on the principle physiological data streams to obtain comprehensive physiological indexes;

[0007] Step S2: Predict individualized physiological indicators based on the comprehensive physiological indicators, and predict the probability of the patient's condition worsening to obtain condition worsening probability data; predict the nursing level demand based on the prediction results of the individualized physiological indicators to obtain nursing level demand probability data; generate a condition trend prediction vector for the condition worsening probability data and the nursing level demand probability data to obtain a condition trend prediction vector;

[0008] Step S3: setting a dynamic warning threshold according to the disease trend prediction vector and the preset warning rules to obtain a dynamic threshold matrix; triggering a warning signal according to the disease trend prediction vector and the dynamic threshold matrix, and determining the warning level to obtain warning level data;

[0009] Step S4: recommending nursing resources based on the disease trend prediction vector and the warning level data to obtain a recommended nursing plan;

[0010] Step S5: recommend a nursing plan to track changes in physiological indicators and obtain a trajectory of changes in physiological indicators; perform nursing effect evaluation based on the trajectory of changes in physiological indicators and obtain nursing effect indicators to achieve the digital management task of graded nursing in the intensive care unit.

[0011] The present invention realizes a comprehensive and accurate quantitative evaluation of the patient's physiological state by real-time acquisition of multi-source heterogeneous physiological data and calculation of the integrated physiological index (IPI) using a deep learning model, laying a solid data foundation for subsequent personalized prediction and accurate early warning. The patient's future physiological indicators are predicted by an individualized physiological model, and the disease trend prediction vector (PTV) is generated in combination with the probability of disease deterioration and the prediction of nursing level requirements, realizing personalized and forward-looking prediction of the patient's disease development trend, providing key information for early intervention and resource allocation. Based on PTV and preset rules, the early warning threshold is dynamically set, and the comprehensive risk score is calculated in combination with the patient's historical data and risk factor weights, realizing an intelligent and hierarchical early warning mechanism, which can more accurately identify high-risk patients. Taking into account the patient's disease trend, individual characteristics and available resources, intelligently recommend personalized nursing plans, improve the accuracy and feasibility of nursing plans, and ensure that patients receive the most appropriate care. By continuously tracking the changes in the patient's physiological indicators and the execution of nursing, the nursing effect is evaluated in real time, and the nursing plan is dynamically adjusted according to the evaluation results, the closed-loop management of the nursing process is realized, the nursing strategy can be continuously optimized, and the nursing quality and patient safety are ultimately improved. Therefore, the present invention provides a digital management method for graded nursing in intensive care units based on the Internet of Things. Through individualized modeling, multi-dimensional data fusion, deep learning and dynamic threshold setting technologies, more accurate disease trend prediction and graded early warning are achieved, thereby better assisting medical staff in providing graded nursing for patients and improving the management level of intensive care units.

[0012] Preferably, step S1 comprises the following steps:

[0013] Step S11: real-time collection of heterogeneous data of the patient through nursing monitoring equipment to obtain a principle physiological data stream;

[0014] Step S12: Associating the original physiological data stream with the patient identity to obtain patient identity associated data; performing physiological data synchronization processing according to the patient identity associated data to obtain patient synchronized physiological data;

[0015] Step S13: extracting physiological data features from the patient's synchronous physiological data to obtain a physiological feature vector;

[0016] Step S14: Calculate the comprehensive physiological index according to the physiological characteristic vector to obtain the comprehensive physiological index.

[0017] The present invention collects heterogeneous data of patients in real time through a variety of nursing monitoring equipment and sensors, which can comprehensively capture changes in the patient's physiological state, avoid information omissions, and provide a richer data basis for subsequent analysis and decision-making. Real-time data acquisition can also timely reflect the dynamic changes of the patient's condition, which is helpful for early detection of potential risks. The original physiological data stream is associated with the patient's identity and time synchronization is processed to ensure the accuracy and consistency of the data and avoid data confusion and errors. Time synchronization processing can align data with different sampling frequencies, facilitate subsequent feature extraction and analysis, and improve the efficiency and accuracy of data processing. Extracting clinically significant features from synchronized physiological data can convert complex waveform data and numerical data into more representative feature vectors, reduce data dimensions, reduce the amount of calculation, and improve the efficiency and accuracy of subsequent analysis. The extracted features can more effectively reflect the changes in the patient's physiological state and provide a more reliable basis for disease assessment and prediction. By calculating the integrated physiological index (IPI) through a deep learning model, multi-dimensional physiological features can be integrated into a single indicator to more comprehensively reflect the patient's overall physiological state. The calculation of IPI takes into account the correlation and weight between various indicators, which can more accurately assess the patient's condition and provide a more reliable basis for subsequent disease prediction and early warning. The use of deep learning models can capture complex nonlinear relationships and reflect the patient's physiological state more accurately than traditional linear models.

[0018] Preferably, step S2 comprises the following steps:

[0019] Step S21: constructing a time series of the comprehensive physiological index to obtain a time series of the comprehensive physiological index;

[0020] Step S22: performing individualized physiological pattern learning on the comprehensive physiological index time series to obtain an individualized physiological model;

[0021] Step S23: using the individualized physiological model to predict individualized physiological indicators to obtain individualized physiological indicator prediction results;

[0022] Step S24: predicting the probability of worsening of the patient's condition based on the prediction results of the individualized physiological indicators to obtain data on the probability of worsening of the condition;

[0023] Step S25: Predicting nursing level requirements based on individualized physiological index prediction results and preset nursing level classification rules to obtain nursing level requirement probability data;

[0024] Step S26: Generate a disease trend prediction vector based on the disease deterioration probability data and the nursing level requirement probability data to obtain a disease trend prediction vector.

[0025] The present invention can retain the information of IPI data changing over time by constructing IPI data into a time series, and provide a basis for subsequent time series analysis and prediction. Time series data can reflect the dynamic change trend of the patient's physiological state, which helps to more accurately predict the development of the disease. The LSTM model is used to learn the individualized physiological pattern of the IPI time series, which can capture the physiological change law of individual patient specificity. Compared with the general model, the individualized model can more accurately reflect the unique physiological pattern of each patient, thereby improving the accuracy and personalization of the prediction. The individualized physiological model is used to predict the future IPI value, which can more accurately predict the future physiological state change trend of the patient. By analyzing the IPI prediction results and predicting the probability of the patient's condition worsening, the potential risk of condition worsening can be warned in advance, so that medical staff can take intervention measures in time to reduce the possibility of condition worsening. According to the IPI prediction results and the preset nursing level division rules, the probability of the patient's demand for different nursing levels is predicted, which helps to plan and allocate nursing resources in advance, optimize the allocation of nursing resources, avoid resource waste, and ensure that patients receive timely and effective care. Integrating the probability of disease worsening and the probability of care level requirement to generate a disease trend prediction vector (PTV) can provide a more comprehensive and integrated disease prediction result, provide a more reliable basis for subsequent early warning triggering and care plan recommendation, and help improve the accuracy of nursing decisions.

[0026] Preferably, step S25 comprises the following steps:

[0027] Step S251: Predicting the change trend of physiological indicators based on the individualized physiological indicator prediction results to obtain a nursing demand trend prediction sequence;

[0028] Step S252: performing dynamic and static nursing level matching according to the nursing demand trend prediction sequence and the preset nursing level division rules to obtain a nursing demand matching result, wherein the preset nursing level division rules include static division rules and dynamic division rules;

[0029] Step S253: constructing a nursing demand probability model according to the nursing demand matching result to obtain a nursing demand probability model; calculating the nursing demand probability according to the nursing demand probability model and the nursing demand matching result to obtain a nursing demand probability matrix;

[0030] Step S254: Obtain nursing resource configuration data; optimize the nursing resource demand probability matrix according to the nursing resource configuration data to obtain nursing level demand probability data.

[0031] The present invention obtains a nursing demand trend prediction sequence by performing trend analysis on the individualized physiological index prediction results, and can more accurately predict the changing trend of the patient's future nursing needs, rather than relying solely on the current physiological index value, so as to provide more accurate nursing management for preparing and adjusting the nursing plan in advance. Combined with static partitioning rules (based on IPI absolute value) and dynamic partitioning rules (based on IPI change rate) for nursing level matching, the patient's condition and changing trend can be considered more comprehensively, avoiding the limitations of a single rule, thereby improving the accuracy and reliability of nursing level assessment. By constructing a nursing demand probability model (such as GMM) and calculating the nursing demand probability matrix, the prediction results of nursing demand can be expressed in the form of probability, more accurately quantifying the patient's demand for different nursing levels, and providing more refined decision support for the optimal configuration of nursing resources. By constructing a nursing demand probability model (such as GMM) and calculating the nursing demand probability matrix, the prediction results of nursing demand can be expressed in the form of probability, more accurately quantifying the patient's demand for different nursing levels, and providing more refined decision support for the optimal configuration of nursing resources. Optimizing the nursing demand probability matrix based on nursing resource configuration data can incorporate the availability of nursing resources into the demand forecast of nursing levels, making the forecast results more in line with actual conditions and avoiding the situation where patient care needs cannot be met due to insufficient resources, thereby improving the utilization efficiency of nursing resources and ensuring that patients receive the most appropriate level of care.

[0032] Preferably, step S252 is specifically as follows:

[0033] Perform point-by-point matching on the nursing demand trend prediction sequence and the static division rule to obtain point-by-point matching results; perform preliminary marking based on the point-by-point matching results to obtain static matching results;

[0034] Extract trend features from the nursing demand trend prediction sequence to obtain trend features; perform dynamic rule matching on the trend features and dynamic division rules to obtain dynamic matching results;

[0035] The static matching results and the dynamic matching results are fused, and matching conflicts are eliminated to obtain fused matching results.

[0036] The nursing needs matching result is generated according to the fusion matching result to obtain the nursing needs matching result.

[0037] The present invention uses static division rules to make a preliminary assessment of the nursing needs at each time point, provides a baseline level of nursing needs, and provides a reference for subsequent dynamic adjustments. By extracting trend features and applying dynamic division rules, it is possible to capture the dynamic changes in nursing needs and make more detailed adjustments to the nursing level, thereby more accurately reflecting the actual development trend of the patient's condition. By fusing the static matching results and the dynamic matching results and performing conflict elimination processing, it is possible to comprehensively consider the patient's baseline nursing needs and dynamic change trends, avoid the limitations of a single rule, and obtain more reliable nursing level assessment results. Giving priority to dynamic matching results highlights the importance of the trend of disease changes and helps to adjust nursing strategies in a timely manner. The nursing needs matching results finally generated combine static and dynamic evaluations, provide a more comprehensive and accurate nursing level prediction, and provide a more reliable basis for subsequent nursing resource recommendations and nursing plan adjustments.

[0038] Preferably, step S3 comprises the following steps:

[0039] Step S31: setting a dynamic warning threshold according to the disease trend prediction vector and the preset warning rules to obtain a dynamic threshold matrix;

[0040] Step S32: performing single-index avoidance comparison on the disease trend prediction vector and the dynamic threshold matrix to obtain a single-index comparison result; performing multi-index fusion on the single-index comparison result to obtain a multi-index fusion result;

[0041] Step S33: Perform early warning trigger judgment according to the multi-indicator fusion result and the dynamic threshold matrix to obtain an early warning trigger signal;

[0042] Step S34: Calculate the risk factor weights according to the disease trend prediction vector, the patient's historical physiological data, and the preset warning classification results to obtain a risk factor weight table;

[0043] Step S35: Calculate the risk score according to the risk factor weight table and the early warning trigger signal to obtain a comprehensive risk score;

[0044] Step S36: Classify the warning level according to the comprehensive risk score and the preset warning classification results to obtain warning level data.

[0045] The present invention avoids the limitation of fixed threshold value by dynamically adjusting the early warning threshold value according to the disease trend prediction vector, can carry out personalized early warning according to the individual situation of the patient, improve the sensitivity and accuracy of early warning, and reduce the occurrence of false alarm and missed alarm. Single index comparison is first carried out, and then multi-index fusion is carried out, so that the patient's disease risk can be more comprehensively evaluated. Multi-index fusion takes into account the comprehensive effect of multiple risk factors, avoids the one-sidedness of a single indicator, and thus improves the reliability of early warning. Early warning trigger judgment is carried out according to the multi-index fusion result and the dynamic threshold matrix, and it can more accurately judge whether the patient needs intervention, and timely trigger the early warning, and provide timely risk prompts for medical staff, so that medical staff can take intervention measures as soon as possible. By combining the disease trend prediction vector, the patient's historical physiological data and the preset early warning classification results to calculate the risk factor weight, it is possible to more accurately evaluate the degree of influence of different risk factors on the patient's condition, thereby providing a more scientific basis for risk scoring. According to the risk factor weight table, the comprehensive risk score can be calculated, and the influence of multiple risk factors can be combined to quantify the overall risk level of the patient, providing a more accurate basis for the early warning level division. By dividing the warning levels according to the comprehensive risk score, the patient's risk level can be graded so that medical staff can take appropriate intervention measures according to different warning levels, thereby improving the pertinence and effectiveness of care.

[0046] Preferably, step S31 includes the following steps:

[0047] Step S311: Acquire the patient's historical physiological data; construct an early warning Bayesian network based on the patient's historical physiological data and the disease trend prediction vector to obtain an early warning Bayesian network;

[0048] Step S312: using the patient's historical physiological data to set the probability distribution of the early warning Bayesian network, and obtaining a probabilistic Bayesian network;

[0049] Step S313: extracting the patient's current physiological index from the principle physiological data stream to obtain the patient's current physiological index;

[0050] Step S314: input the patient's current physiological indicators and the disease trend prediction vector into the probabilistic Bayesian network to perform dynamic threshold calculation to obtain a dynamic threshold matrix.

[0051] The present invention constructs a Bayesian network by using the patient's historical physiological data and the disease trend prediction vector, which can effectively express and learn the complex probability relationship between different physiological indicators, risk factors and disease trends, and provide a more explanatory and reasoning model for dynamic threshold setting. The conditional probability distribution of the Bayesian network is set using the patient's historical physiological data, and the individual differences and historical information of the patient can be integrated into the model, so that the model is more in line with the patient's actual situation, thereby improving the accuracy and personalization of dynamic threshold setting. The patient's current physiological indicators are extracted in real time, and the patient's latest physiological state changes can be captured, providing real-time input data for dynamic threshold calculation, ensuring that the warning threshold can be dynamically adjusted according to the patient's disease changes. The patient's current physiological indicators and disease trend prediction vector are input into the Bayesian network with probability for reasoning, and the warning threshold can be dynamically calculated according to the patient's real-time state and predicted trend, thereby improving the sensitivity and accuracy of the warning, and better adapting to the dynamic changes of the patient's disease. Using the Bayesian network for reasoning can effectively utilize prior knowledge and real-time data to improve the reliability of threshold setting.

[0052] Preferably, step S311 is specifically as follows:

[0053] Perform data preprocessing on the patient's historical physiological data and disease trend prediction vector to obtain preprocessed data;

[0054] An initial node set is defined according to the preprocessed data to obtain an initial node set; node similarity is calculated according to the preprocessed data and the initial node set to obtain a node similarity matrix;

[0055] Perform node clustering analysis on the node similarity matrix to obtain node clustering results; construct a hierarchical structure based on the node clustering results and the initial node set to obtain a node hierarchy;

[0056] Using the preprocessed data to learn the dependency relationship of the node hierarchy, a node hierarchy with dependency relationship is obtained;

[0057] A Bayesian network is constructed according to the node hierarchy with dependency relationships to obtain an early warning Bayesian network.

[0058] The present invention can improve the quality and consistency of data through data preprocessing steps, including missing value processing, standardization and unique hot encoding, provide a more reliable data basis for subsequent analysis and modeling, and improve the performance and stability of the model. A clearly defined initial node set ensures that the Bayesian network can contain all relevant variables, laying the foundation for building a model that can fully reflect the patient's physiological state and disease trend. The node similarity matrix is ​​calculated to quantify the correlation between different variables, provide a basis for subsequent cluster analysis and hierarchical structure construction, and help to discover the potential relationship between variables. Node cluster analysis can aggregate nodes with high similarity together, simplify the structure of the Bayesian network, and improve the efficiency and interpretability of the model, while helping to discover the intrinsic connection between variables. Constructing a node hierarchy according to the clustering results can reflect the hierarchical relationship between variables, make the structure of the Bayesian network clearer, easier to understand and explain, and help to more effectively express the dependency between variables. Learning the dependency between nodes in a data-driven manner can avoid the subjectivity of artificial settings, and make the structure of the Bayesian network more consistent with the actual distribution of data, thereby improving the accuracy and reliability of the model. The final warning Bayesian network, based on data-driven hierarchical structures and dependencies, can more accurately reflect the complex relationship between patients' physiological states and disease trends, and provide a more reliable model foundation for subsequent dynamic threshold calculations and warnings.

[0059] Preferably, step S4 comprises the following steps:

[0060] Step S41: performing individualized patient information retrieval according to the disease trend prediction vector to obtain a patient feature file;

[0061] Step S42: Acquire the nursing resource status according to the warning level data to obtain a list of available resources;

[0062] Step S43: generating an initial nursing plan according to the warning level data, the patient characteristic file and the available resource list to obtain an initial nursing plan;

[0063] Step S44: The initial nursing plan is adjusted individually according to the patient's characteristic profile and the list of available resources to obtain a recommended nursing plan.

[0064] The present invention can construct a more complete patient characteristic file by retrieving the patient's individualized information, including age, gender, allergy history, past medical history, etc., to provide more comprehensive information support for the subsequent nursing plan formulation, and help to formulate a more personalized nursing plan. According to the warning level, real-time nursing resource status information, such as available beds, medical staff on duty, nursing monitoring equipment, etc., can be obtained to ensure that the recommended nursing plan is feasible in actual operation, avoid the situation where the nursing plan cannot be implemented due to insufficient resources, and improve the utilization efficiency of nursing resources. According to the warning level, the patient characteristic file and the available resource list, the initial nursing plan is generated, and a preliminary nursing plan framework can be quickly provided, which provides a basis for subsequent personalized adjustments and speeds up the formulation process of the nursing plan. According to the patient characteristic file and the available resource list, the initial nursing plan is personalized adjusted to ensure that the recommended nursing plan is more in line with the individual needs and actual situation of the patient, and improves the effectiveness and safety of nursing. Adjustments are made considering resource limitations to ensure the executability of the plan.

[0065] Preferably, step S5 comprises the following steps:

[0066] Step S51: Recording nursing execution status according to the recommended nursing plan to obtain nursing execution record data;

[0067] Step S52: Tracking the changes of physiological indicators according to the recommended nursing plan to obtain the trajectory of physiological indicator changes;

[0068] Step S53: Perform a preliminary evaluation of nursing effects according to the nursing execution record data and the physiological index change trajectory to obtain preliminary effect evaluation data;

[0069] Step S54: dynamically adjusting the nursing plan according to the preliminary effect evaluation data and the recommended nursing plan to obtain an adjusted nursing plan;

[0070] Step S55: Generate nursing effect indicators based on the preliminary effect evaluation data and the adjusted nursing plan to obtain nursing effect indicators.

[0071] The present invention can provide objective data support for nursing effect evaluation by recording the nursing execution situation in detail, including operation content, time, executor and patient response, facilitate tracking the nursing process, and improve the transparency and traceability of nursing work. Continuously tracking the change trajectory of the patient's physiological indicators can dynamically monitor the implementation effect of the nursing plan, provide real-time physiological data basis for nursing effect evaluation, and help to timely discover changes in the condition. According to the nursing execution record and the change trajectory of physiological indicators, a preliminary evaluation can be made, the effectiveness of the nursing plan can be quickly judged, timely feedback can be provided for the dynamic adjustment of the nursing plan, and it is helpful to optimize the nursing strategy. According to the preliminary evaluation results, the nursing plan can be dynamically adjusted, and medical staff can make personalized plan adjustments according to the actual situation and nursing effect of the patient, improve the accuracy and effectiveness of nursing, and ensure that the patient always receives the most appropriate nursing plan. Generate a nursing effect index (CEI), which can quantify the nursing effect, provide objective indicators for nursing quality evaluation, and provide data support for the continuous improvement of nursing plans, and promote the continuous improvement of nursing quality. Using CEI can more effectively compare the effects of different nursing plans and make data-driven nursing decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic diagram of the steps of a digital management method for graded nursing care in an intensive care unit based on the Internet of Things;

[0073] Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart;

[0074] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0075] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0076] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0077] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0078] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0079] To achieve this, please refer to Figures 1 to 3 , a digital management method for graded nursing care in an intensive care unit based on the Internet of Things, comprising the following steps:

[0080] Step S1: real-time collection of heterogeneous data of patients by nursing monitoring equipment to obtain principle physiological data streams; comprehensive physiological index calculations are performed on the principle physiological data streams to obtain comprehensive physiological indexes;

[0081] Step S2: Predict individualized physiological indicators based on the comprehensive physiological indicators, and predict the probability of the patient's condition worsening to obtain condition worsening probability data; predict the nursing level demand based on the prediction results of the individualized physiological indicators to obtain nursing level demand probability data; generate a condition trend prediction vector for the condition worsening probability data and the nursing level demand probability data to obtain a condition trend prediction vector;

[0082] Step S3: setting a dynamic warning threshold according to the disease trend prediction vector and the preset warning rules to obtain a dynamic threshold matrix; triggering a warning signal according to the disease trend prediction vector and the dynamic threshold matrix, and determining the warning level to obtain warning level data;

[0083] Step S4: recommending nursing resources based on the disease trend prediction vector and the warning level data to obtain a recommended nursing plan;

[0084] Step S5: recommend a nursing plan to track changes in physiological indicators and obtain a trajectory of changes in physiological indicators; perform nursing effect evaluation based on the trajectory of changes in physiological indicators and obtain nursing effect indicators to achieve the digital management task of graded nursing in the intensive care unit.

[0085] In the embodiment of the present invention, reference Figure 1 As shown, it is a schematic diagram of the steps of the digital management method for graded nursing care in intensive care units based on the Internet of Things of the present invention. In this example, the digital management method for graded nursing care in intensive care units based on the Internet of Things includes the following steps:

[0086] Step S1: real-time collection of heterogeneous data of patients by nursing monitoring equipment to obtain principle physiological data streams; comprehensive physiological index calculations are performed on the principle physiological data streams to obtain comprehensive physiological indexes;

[0087] In an embodiment of the present invention, by connecting a variety of nursing monitoring devices and sensors (such as ECG sensors, sphygmomanometers, oximeters, smart mattresses, bedside monitors, environmental sensors, etc.) of the patient, a variety of physiological data and environmental data of the patient are collected in real time to form a raw physiological data stream (RPDS). Using the patient's unique RFID identifier, the data from different devices are associated with the patient's identity, and time synchronization is performed to generate patient synchronized physiological data (PSPD). Then, physiological features such as heart rate variability, blood pressure characteristics, blood oxygen saturation characteristics, and respiratory characteristics are extracted from the PSPD to form a physiological feature vector (PFV). Finally, the PFV is calculated using a pre-trained deep learning model (such as an MLP-based model) to obtain a comprehensive physiological index (IPI), which is stored in the time series database Influ×DB.

[0088] Step S2: Predict individualized physiological indicators based on the comprehensive physiological indicators, and predict the probability of the patient's condition worsening to obtain condition worsening probability data; predict the nursing level demand based on the prediction results of the individualized physiological indicators to obtain nursing level demand probability data; generate a condition trend prediction vector for the condition worsening probability data and the nursing level demand probability data to obtain a condition trend prediction vector;

[0089] In an embodiment of the present invention, the IPI data of the patient in the past 24 hours is read from Influ×DB to construct an IPI time series (IPITS). The IPITS is trained using the LSTM model to learn the individualized physiological patterns of the patient and obtain an individualized physiological model (IPM). The IPM is used to predict the IPI value for the next 12 hours. Based on the IPI prediction results, the probability of disease deterioration (POD) is calculated, and the probability of demand for different levels of care in the next 12 hours is predicted based on the preset nursing level division rules and the IPI change rate. Finally, the POD and the nursing level demand probability are integrated together to generate a disease trend prediction vector (PTV).

[0090] Step S3: setting a dynamic warning threshold according to the disease trend prediction vector and the preset warning rules to obtain a dynamic threshold matrix; triggering a warning signal according to the disease trend prediction vector and the dynamic threshold matrix, and determining the warning level to obtain warning level data;

[0091] In an embodiment of the present invention, the warning threshold is dynamically adjusted according to the PTV and the preset warning rules to form a dynamic threshold matrix (DTM). Each indicator in the PTV is compared with the corresponding threshold in the DTM, and the comparison result is logically ANDed to determine whether the warning is triggered. If the warning is triggered, the risk factor weight is calculated according to the PTV, the patient's historical physiological data and the preset warning classification result, and the comprehensive risk score is calculated according to the risk factor weight and the warning trigger signal. Finally, the warning level is determined according to the comprehensive risk score and the preset warning classification result.

[0092] Step S4: recommending nursing resources based on the disease trend prediction vector and the warning level data to obtain a recommended nursing plan;

[0093] In an embodiment of the present invention, the individualized information of the patient is retrieved from the electronic medical record system according to the PTV to form a patient characteristic file. The available nursing resource status information is obtained from the hospital resource management system according to the warning level to form an available resource list. According to the warning level, the patient characteristic file and the available resource list, a matching initial nursing plan is retrieved from a predefined nursing plan template library. Then, according to the patient characteristic file and the available resource list, the medical staff makes personalized adjustments to the initial nursing plan, and finally obtains a recommended nursing plan.

[0094] Step S5: recommending a nursing plan to track changes in physiological indicators and obtain a trajectory of changes in physiological indicators; performing nursing effect evaluation based on the trajectory of changes in physiological indicators and obtaining nursing effect indicators to achieve digital management tasks for graded nursing in intensive care units;

[0095] In an embodiment of the present invention, medical staff records the execution of the recommended nursing plan through a mobile nursing terminal or an electronic nursing record system to form nursing execution record data. The system continuously collects the patient's physiological index data, and records the physiological index data before and after the nursing execution according to the monitoring frequency specified in the recommended nursing plan to form a physiological index change trajectory. Based on the nursing execution record data and the physiological index change trajectory, a preliminary evaluation of the nursing effect is performed. Based on the preliminary evaluation results and the recommended nursing plan, the nursing plan is dynamically adjusted. Finally, a nursing effect index (CEI) is generated based on the preliminary evaluation results and the adjusted nursing plan.

[0096] Preferably, step S1 comprises the following steps:

[0097] Step S11: real-time collection of heterogeneous data of the patient through nursing monitoring equipment to obtain a principle physiological data stream;

[0098] Step S12: Associating the original physiological data stream with the patient identity to obtain patient identity associated data; performing physiological data synchronization processing according to the patient identity associated data to obtain patient synchronized physiological data;

[0099] Step S13: extracting physiological data features from the patient's synchronous physiological data to obtain a physiological feature vector;

[0100] Step S14: Calculate the comprehensive physiological index according to the physiological characteristic vector to obtain the comprehensive physiological index.

[0101] In an embodiment of the present invention, in the intensive care unit, real-time data collection is performed on the patient through a variety of nursing monitoring devices and sensors deployed. For example, an ECG sensor attached to the patient's chest collects electrocardiogram data at a frequency of 250Hz; a non-invasive sphygmomanometer connected to the patient's arm collects blood pressure data once a minute; a pulse oximeter collects blood oxygen saturation and pulse rate data once a second; and a smart mattress sensor monitors the patient's body position and respiratory rate in real time. The bedside monitor transmits comprehensive data including heart rate, respiratory rate, blood pressure and blood oxygen saturation every 5 seconds through a standard interface (such as HL7). The ward environment temperature and humidity sensor collects ambient temperature and humidity data once a minute. All collected data are time-stamped and transmitted to the central data processing server in real time through the wireless network in the ward to form a raw physiological data stream (RPDS). The RPDS data stream format adopts JSON format and is stored in the Kafka distributed message queue for subsequent processing.

[0102] Read RPDS data from the Kafka message queue. Each data packet contains a unique RFID identifier for the patient. The system associates data from different sensors and nursing monitoring devices with the corresponding patient identity through the RFID identifier to generate patient identity-associated data. Considering the different sampling frequencies of different devices, the system uses a timestamp alignment algorithm to synchronize patient identity-associated data. For example, taking 1 minute as a time window, all data collected from the same patient within the time window are aggregated. If the data of a sensor is missing within the time window, it is filled with the data from the previous time window. After synchronization, patient synchronized physiological data (PSPD) is generated, which contains multi-dimensional physiological data of each patient in each time window. PSPD data is stored in the Hadoop Distributed File System (HDFS) in Parquet format.

[0103] Read PSPD data from HDFS. Perform feature extraction on the patient's synchronized physiological data within each time window. For example, use wavelet transform to extract heart rate variability (HRV) features from ECG data; use peak detection algorithm to extract systolic blood pressure, diastolic blood pressure and mean arterial pressure from blood pressure data; use statistical methods to extract the mean, standard deviation and coefficient of variation of blood oxygen saturation from blood oxygen saturation data; use fast Fourier transform to extract respiratory frequency and amplitude features from respiratory signals. All extracted features are combined into a numerical vector, namely physiological feature vector (PFV). PFV data is stored in HBase database with patient ID and time window as primary key.

[0104] Read PFV data from the HBase database. Use a pre-trained deep learning model (e.g., a model based on a multi-layer perceptron (MLP)) to calculate the integrated physiological index (IPI) of PFV. The input of the model is PFV, and the output is a value between 0 and 1, representing the overall physiological state of the patient. For example, when PFV is input into the trained MLP model, the model output is 0.85, indicating that the patient's current physiological state is good. The IPI data is stored in the time series database Influ×DB together with the corresponding timestamp for subsequent trend analysis.

[0105] Preferably, step S2 comprises the following steps:

[0106] Step S21: constructing a time series of the comprehensive physiological index to obtain a time series of the comprehensive physiological index;

[0107] Step S22: performing individualized physiological pattern learning on the comprehensive physiological index time series to obtain an individualized physiological model;

[0108] Step S23: using the individualized physiological model to predict individualized physiological indicators to obtain individualized physiological indicator prediction results;

[0109] Step S24: predicting the probability of worsening of the patient's condition based on the prediction results of the individualized physiological indicators to obtain data on the probability of worsening of the condition;

[0110] Step S25: Predicting nursing level requirements based on individualized physiological index prediction results and preset nursing level classification rules to obtain nursing level requirement probability data;

[0111] Step S26: Generate a disease trend prediction vector based on the disease deterioration probability data and the nursing level requirement probability data to obtain a disease trend prediction vector.

[0112] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0113] Step S21: constructing a time series of the comprehensive physiological index to obtain a time series of the comprehensive physiological index;

[0114] In an embodiment of the present invention, the IPI data of a specified patient in the past 24 hours is read from the Influ×DB database and arranged in timestamp order to generate the IPI time series (IPITS) of the patient. The sampling frequency of the time series is once per minute. If the data is missing, linear interpolation is used to fill it. For example, if the patient ID is "12345", the IPI value of the patient every minute in the past 24 hours is queried from Influ×DB, and an IPI time series containing 1440 data points is generated and stored as a NumPy array.

[0115] Step S22: performing individualized physiological pattern learning on the comprehensive physiological index time series to obtain an individualized physiological model;

[0116] In an embodiment of the present invention, the IPITS generated in step S21 is trained using a long short-term memory network (LSTM) model to learn the individualized physiological patterns of patients. The LSTM model comprises two LSTM layers, each of which comprises 64 hidden units, followed by a fully connected layer, and the output dimension is 1. The Adam optimizer is used for training, the learning rate is set to 0.001, the batch size is set to 32, and the number of training rounds is set to 100. The loss function is the mean square error (MSE). The training process uses an early stopping strategy to prevent overfitting. After the training is completed, the trained LSTM model is saved as an individualized physiological model (IPM).

[0117] Step S23: using the individualized physiological model to predict individualized physiological indicators to obtain individualized physiological indicator prediction results;

[0118] In the embodiment of the present invention, the IPM trained in step S22 is used to predict the IPI for the next 12 hours. The IPI data for the past 12 hours is used as the model input to predict the IPI value for the next 12 hours, and a prediction sequence containing 720 data points is generated. For example, the IPI data for the past 12 hours of patient "12345" is input into its corresponding IPM, and the model outputs the predicted IPI value for the next 12 hours.

[0119] Step S24: predicting the probability of worsening of the patient's condition based on the prediction results of the individualized physiological indicators to obtain data on the probability of worsening of the condition;

[0120] In an embodiment of the present invention, the probability of worsening of the patient's condition is calculated based on the IPI prediction results. The IPI threshold is preset to 0.3. If any of the predicted IPI values ​​in the next 12 hours is less than or equal to 0.3, it is considered that the patient's condition may worsen. The ratio of the number of time points with IPI values ​​less than or equal to 0.3 to the total predicted time points in the next 12 hours is calculated as the probability of worsening of the condition (POD). For example, if patient "12345" has 120 time points with IPI predicted values ​​less than or equal to 0.3 in the next 12 hours, then POD = 120 / 720 = 0.167.

[0121] Step S25: Predicting nursing level requirements based on individualized physiological index prediction results and preset nursing level classification rules to obtain nursing level requirement probability data;

[0122] In an embodiment of the present invention, the probability of a patient's need for different nursing levels in the next 12 hours is predicted based on the IPI prediction results and the preset nursing level classification rules. The preset nursing level classification rules are as follows: IPI greater than or equal to 0.7 is first-level nursing, IPI between 0.4 and 0.7 is second-level nursing, and IPI less than 0.4 is third-level nursing. The ratio of the number of time points belonging to each nursing level to the total predicted time points in the next 12 hours is calculated as the probability of need for the nursing level. For example, if patient "12345" has 360 time points with an IPI prediction value greater than or equal to 0.7 in the next 12 hours, the probability of first-level nursing need is 360 / 720=0.5.

[0123] Step S26: Generate a disease trend prediction vector based on the disease deterioration probability data and the nursing level requirement probability data to obtain a disease trend prediction vector;

[0124] In the embodiment of the present invention, the probability of disease progression (POD) and the probability of nursing level requirement calculated in step S25 are integrated together to generate a disease trend prediction vector (PTV). PTV is a four-dimensional vector, which represents the probability of disease progression, the probability of primary nursing requirement, the probability of secondary nursing requirement and the probability of tertiary nursing requirement. For example, the PTV of patient "12345" is [0.167, 0.5, 0.3, 0.033].

[0125] Preferably, step S25 comprises the following steps:

[0126] Step S251: Predicting the change trend of physiological indicators based on the individualized physiological indicator prediction results to obtain a nursing demand trend prediction sequence;

[0127] Step S252: Perform static and dynamic nursing level matching based on the nursing demand trend prediction sequence and the preset nursing level division rules to obtain the nursing demand matching result, where the preset nursing level division rules include static division rules and dynamic division rules;

[0128] Step S253: Construct a nursing demand probability model based on the nursing demand matching result to obtain the nursing demand probability model; calculate the nursing demand probability according to the nursing demand probability model and the nursing demand matching result to obtain the nursing demand probability matrix;

[0129] Step S254: Obtain the nursing resource allocation data; optimize the nursing demand probability matrix according to the nursing resource allocation data to obtain the nursing level demand probability data.

[0130] In the embodiment of the present invention, based on the individual physiological index prediction result (the IPI prediction sequence for the next 12 hours), the Savitzky-Golay filter is used to smooth the IPI prediction sequence to remove noise and highlight the trend. Then, the change amount of the IPI value at adjacent time points is calculated to obtain the IPI change rate sequence. For example, if the IPI prediction value at the i-th time point is 0.6 and the IPI prediction value at the i + 1-th time point is 0.55, then the IPI change rate at the i-th time point is 0.55 - 0.6 = -0.05. The IPI change rate sequence is used as the nursing demand trend prediction sequence. A positive value indicates an improving trend of the condition, a negative value indicates a deteriorating trend of the condition, and a zero value indicates a stable condition.

[0131] Static division rule: Divide the nursing level according to the absolute value of IPI. When IPI ≥ 0.7, it is first-level nursing; when 0.4 ≤ IPI < 0.7, it is second-level nursing; when IPI < 0.4, it is third-level nursing. Dynamic division rule: Divide the nursing level according to the IPI change rate. When the IPI change rate ≤ -0.05, it is third-level nursing; when -0.05 < IPI change rate < 0.02, it is second-level nursing; when the IPI change rate ≥ 0.02, it is first-level nursing. Match the nursing demand trend prediction sequence with the static division rule and the dynamic division rule respectively. For example, if the IPI prediction value at a certain time point is 0.5 and the IPI change rate is -0.06, the static matching result is second-level nursing and the dynamic matching result is third-level nursing. Store the static matching results and dynamic matching results at all time points as two sequences respectively. Subsequently, conflict resolution is performed on the two sequences. Give priority to the dynamic matching result, that is, if the static matching result and the dynamic matching result are inconsistent, the dynamic matching result is adopted. The result after conflict resolution is used as the final nursing demand matching result.

[0132] The Gaussian mixture model (GMM) was used to model the nursing needs matching results and obtain the nursing needs probability model. The number of components of the GMM was set to 3, corresponding to three nursing levels. The expectation maximization (EM) algorithm was used to train the GMM to learn the probability distribution of each nursing level. After the training was completed, the nursing needs matching results were input into the trained GMM model, and the probability of each time point belonging to each nursing level was calculated to obtain a 720×3 nursing needs probability matrix. Each row of the matrix represents a time point, and each column represents a nursing level.

[0133] The currently available nursing resource configuration data is obtained from the hospital information system, including the number of available beds for each nursing level, the number of medical staff, etc. For example, the number of available beds for primary nursing is 10, the number of available beds for secondary nursing is 20, and the number of available beds for tertiary nursing is 5. According to the nursing resource configuration data, the nursing demand probability matrix is ​​adjusted. For example, if the probability of tertiary nursing demand is high at a certain point in time, but the number of available beds for tertiary nursing is full, the probability of tertiary nursing demand at that point in time is reduced, and the excess probability is proportionally distributed to the probabilities of primary and secondary nursing demand. Finally, the optimized nursing level demand probability data is obtained, which is still a 720×3 matrix, but takes into account the limitation of nursing resources.

[0134] Preferably, step S252 is specifically as follows:

[0135] Perform point-by-point matching on the nursing demand trend prediction sequence and the static division rule to obtain point-by-point matching results; perform preliminary marking based on the point-by-point matching results to obtain static matching results;

[0136] Extract trend features from the nursing demand trend prediction sequence to obtain trend features; perform dynamic rule matching on the trend features and dynamic division rules to obtain dynamic matching results;

[0137] The static matching results and the dynamic matching results are fused, and matching conflicts are eliminated to obtain fused matching results.

[0138] The nursing needs matching result is generated according to the fusion matching result to obtain the nursing needs matching result.

[0139] In an embodiment of the present invention, a nursing demand trend prediction sequence (i.e., an IPI change rate sequence) and a static division rule (the correspondence between the IPI absolute value and the nursing level) are obtained. The IPI absolute value at each time point is compared with the static division rule to determine the nursing level corresponding to the time point. For example, if the IPI absolute value is 0.8, it matches the first-level nursing, if the IPI absolute value is 0.5, it matches the second-level nursing, and if the IPI absolute value is 0.2, it matches the third-level nursing. The matching results at each time point are recorded to form a sequence with a length of 720, i.e., the point-by-point matching results. The point-by-point matching results are used as static matching results, and the numbers 1, 2, and 3 are used to represent the first-level, second-level, and third-level nursing, respectively.

[0140] Get the nursing demand trend forecast sequence (i.e., IPI change rate sequence). Use the sliding window method to extract trend features. Set the window size to 5, that is, use the IPI change rate of 5 consecutive time points to calculate the trend features of the current time point. Calculate the mean and standard deviation of the IPI change rate within the window. Take the mean as the trend strength and the standard deviation as the trend volatility. For example, if the IPI change rate of 5 consecutive time points is [-0.02, -0.03, -0.04, -0.05, -0.06], the trend strength is -0.04 and the trend volatility is 0.014. Combine the trend strength and volatility of each time point into a two-dimensional vector as the trend feature of that time point. Then, match the trend features of each time point with the dynamic partitioning rules. For example, if the trend strength is less than -0.05, it matches tertiary care; if the trend strength is between -0.05 and 0.02, it matches secondary care; if the trend strength is greater than 0.02, it matches primary care. The matching results at each time point are recorded to form a sequence of length 720, namely the dynamic matching result. The numbers 1, 2, and 3 are used to represent primary, secondary, and tertiary care, respectively.

[0141] Compare the static matching result and the dynamic matching result. If the two results are the same, use the result directly. If the two results are different, use the dynamic matching result first. For example, if the static matching result at a certain time point is level 2 care and the dynamic matching result is level 3 care, the final result is level 3 care. Record the final matching result at each time point to form a sequence of length 720, i.e., the fused matching result, with the numbers 1, 2, and 3 representing level 1, level 2, and level 3 care, respectively.

[0142] The fusion matching results are used as the final care demand matching results. This sequence contains the predicted care level at each time point as the input for subsequent steps. For example, the final generated care demand matching result sequence may be similar to [1, 1, 2, 2, 3, 3, 2, 1, ...], where each number represents the care level at the corresponding time point.

[0143] Preferably, step S3 comprises the following steps:

[0144] Step S31: setting a dynamic warning threshold according to the disease trend prediction vector and the preset warning rules to obtain a dynamic threshold matrix;

[0145] Step S32: performing single-index avoidance comparison on the disease trend prediction vector and the dynamic threshold matrix to obtain a single-index comparison result; performing multi-index fusion on the single-index comparison result to obtain a multi-index fusion result;

[0146] Step S33: Perform early warning trigger judgment according to the multi-indicator fusion result and the dynamic threshold matrix to obtain an early warning trigger signal;

[0147] Step S34: Calculate the risk factor weights according to the disease trend prediction vector, the patient's historical physiological data, and the preset warning classification results to obtain a risk factor weight table;

[0148] Step S35: Calculate the risk score according to the risk factor weight table and the early warning trigger signal to obtain a comprehensive risk score;

[0149] Step S36: Classify the warning level according to the comprehensive risk score and the preset warning classification results to obtain warning level data.

[0150] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0151] Step S31: setting a dynamic warning threshold according to the disease trend prediction vector and the preset warning rules to obtain a dynamic threshold matrix;

[0152] In an embodiment of the present invention, a prediction trend vector (PTV) of the disease is obtained. The preset warning rules include a probability of disease worsening (POD) threshold, a first-level nursing demand probability threshold, a second-level nursing demand probability threshold, and a third-level nursing demand probability threshold. The basic thresholds are set as: POD>0.2, first-level nursing demand probability<0.6, second-level nursing demand probability>0.3, and third-level nursing demand probability>0.1. The basic thresholds are dynamically adjusted according to the patient's current IPI value. For example, if the patient's current IPI value is lower than 0.5, the POD threshold is reduced to 0.15, and the third-level nursing demand probability threshold is increased to 0.15. The dynamically adjusted thresholds are formed into a 4×1 dynamic threshold matrix (DTM)

[0153] Step S32: performing single-index avoidance comparison on the disease trend prediction vector and the dynamic threshold matrix to obtain a single-index comparison result; performing multi-index fusion on the single-index comparison result to obtain a multi-index fusion result;

[0154] In the embodiment of the present invention, each indicator in the PTV is compared with the corresponding threshold in the DTM to obtain a single indicator comparison result. For example, if the POD in the PTV is 0.25 and the POD threshold in the DTM is 0.15, the single indicator comparison result of the POD is true (True). The single indicator comparison results of all indicators are logically ANDed to obtain a multi-indicator fusion result. For example, if the single indicator comparison results of all indicators are true, the multi-indicator fusion result is true; if the single indicator comparison result of any indicator is false, the multi-indicator fusion result is false.

[0155] Step S33: Perform early warning trigger judgment according to the multi-indicator fusion result and the dynamic threshold matrix to obtain an early warning trigger signal;

[0156] In the embodiment of the present invention, if the multi-indicator fusion result is true, an early warning is triggered, and the early warning trigger signal is 1; otherwise, no early warning is triggered, and the early warning trigger signal is 0.

[0157] Step S34: Calculate the risk factor weights according to the disease trend prediction vector, the patient's historical physiological data, and the preset warning classification results to obtain a risk factor weight table;

[0158] In an embodiment of the present invention, historical physiological data of the patient, such as IPI data of the past 7 days, past medical history, etc., are obtained from a patient database. The preset warning classification results include low risk, medium risk, and high risk. A logistic regression model is used to train a risk factor weight calculation model with PTV, historical physiological data of the patient, and warning classification results as input. The model output is the weight of each risk factor, such as age, past medical history, current IPI value, POD, etc. These weights are stored in a table to form a risk factor weight table.

[0159] Step S35: Calculate the risk score according to the risk factor weight table and the early warning trigger signal to obtain a comprehensive risk score;

[0160] In an embodiment of the present invention, if the early warning trigger signal is 1, the comprehensive risk score is calculated according to the risk factor weight table. The value of each risk factor is multiplied by its corresponding weight, and then all the results are added to obtain a comprehensive risk score. For example, if the age weight is 0.2 and the patient is 60 years old, the age score is 0.2*60=12. All risk factor scores are added to obtain a comprehensive risk score. If the early warning trigger signal is 0, the comprehensive risk score is 0.

[0161] Step S36: Classify the warning level according to the comprehensive risk score and the preset warning classification result to obtain warning level data;

[0162] In the embodiment of the present invention, the comprehensive risk score is divided into warning levels according to the preset warning classification results. For example, a comprehensive risk score below 5 is low risk, corresponding to warning level 1; a comprehensive risk score between 5 and 10 is medium risk, corresponding to warning level 2; a comprehensive risk score above 10 is high risk, corresponding to warning level 3. The divided warning levels are output as warning level data.

[0163] Preferably, step S31 includes the following steps:

[0164] Step S311: Acquire the patient's historical physiological data; construct an early warning Bayesian network based on the patient's historical physiological data and the disease trend prediction vector to obtain an early warning Bayesian network;

[0165] Step S312: using the patient's historical physiological data to set the probability distribution of the early warning Bayesian network, and obtaining a probabilistic Bayesian network;

[0166] Step S313: extracting the patient's current physiological index from the principle physiological data stream to obtain the patient's current physiological index;

[0167] Step S314: input the patient's current physiological indicators and the disease trend prediction vector into the probabilistic Bayesian network to perform dynamic threshold calculation to obtain a dynamic threshold matrix.

[0168] In an embodiment of the present invention, the IPI data, age, gender, and diagnostic information related to the patient's condition in the past 7 days, such as whether there is hypertension, diabetes, etc., are obtained from the database. At the same time, the condition trend prediction vector (PTV) is obtained. Using Python's `pgmpy` library, a Bayesian network is constructed based on the patient's historical physiological data and PTV. The network nodes include: age, gender, hypertension, diabetes, average IPI value (past 7 days), IPI change trend (from PTV), probability of disease worsening (from PTV) and nursing level requirements (from PTV). The hill climbing algorithm is used to learn the dependencies between network nodes and set the network structure. For example, age may affect the probability of hypertension and diabetes; historical average IPI and IPI change trend may affect the probability of disease worsening and nursing level requirements. The final generated Bayesian network is stored in the BayesianModel object format of `pgmpy`.

[0169] Using the patient's historical physiological data, set the conditional probability distribution (CPD) for each node in the Bayesian network. For discrete variables, such as gender, hypertension, and diabetes, use the maximum likelihood estimation method to learn CPD from historical data. For continuous variables, such as IPI, it is assumed to obey a Gaussian distribution, and the parameters (mean and variance) of the Gaussian distribution are estimated using historical data. Add the learned CPD to the Bayesian network to obtain a Bayesian network with probability. For example, based on historical data, set a CPD with a higher probability of worsening in patients with hypertension. The final Bayesian network with probability is still stored in the BayesianModel object format of `pgmpy`, but contains CPD information for each node.

[0170] Get the latest raw physiological data stream (RPDS). Extract the patient's current physiological indicators, including the latest IPI value, heart rate, blood pressure, blood oxygen saturation, etc. Store these indicators in a dictionary, such as `{"IPI": 0.65, "HeartRate": 80, "BloodPressure": "120 / 80", "SpO2": 98}`.

[0171] The patient's current physiological indicators and PTV are input into the probabilistic Bayesian network as evidence. Using a Bayesian network inference algorithm, such as a belief propagation algorithm, the probability distribution of the probability of worsening of the condition, the probability of primary care demand, the probability of secondary care demand, and the probability of tertiary care demand is calculated given the evidence. According to the preset risk level (for example, the false alarm rate is controlled within 5%), the dynamic threshold is determined from the probability distribution. For example, if the probability distribution of the probability of worsening of the condition shows that when the probability is greater than 0.18, the false alarm rate is less than 5%, then 0.18 is used as the dynamic threshold of POD. The four calculated dynamic thresholds are combined into a 4×1 dynamic threshold matrix (DTM).

[0172] Preferably, step S311 is specifically as follows:

[0173] Perform data preprocessing on the patient's historical physiological data and disease trend prediction vector to obtain preprocessed data;

[0174] An initial node set is defined according to the preprocessed data to obtain an initial node set; node similarity is calculated according to the preprocessed data and the initial node set to obtain a node similarity matrix;

[0175] Perform node clustering analysis on the node similarity matrix to obtain node clustering results; construct a hierarchical structure based on the node clustering results and the initial node set to obtain a node hierarchy;

[0176] Using the preprocessed data to learn the dependency relationship of the node hierarchy, a node hierarchy with dependency relationship is obtained;

[0177] A Bayesian network is constructed according to the node hierarchy with dependency relationships to obtain an early warning Bayesian network.

[0178] In an embodiment of the present invention, the IPI data, age, gender, hypertension, diabetes, and disease trend prediction vector (PTV) of the patient in the past 7 days are obtained. Missing values ​​are processed for IPI data, and missing values ​​are filled using linear interpolation. Continuous variables, such as age and IPI data, are standardized and converted into data with a mean of 0 and a standard deviation of 1. Categorical variables, such as gender, hypertension, and diabetes, are uniquely encoded. The preprocessed data is stored in a PandasDataFrame as preprocessed data.

[0179] Based on the variables contained in the preprocessed data, the initial node set is defined. The node set includes: age, gender, hypertension, diabetes, average IPI (past 7 days), IPI change trend (from PTV), probability of disease progression (from PTV), and care level requirements (from PTV). Each node represents a variable.

[0180] Use the Pearson correlation coefficient to calculate the similarity between the variables in the preprocessed data. For continuous variables, directly calculate the Pearson correlation coefficient. For categorical variables, convert the one-hot encoded vector into a numerical value and then calculate the Pearson correlation coefficient. Store the calculated correlation coefficients in a matrix to form a node similarity matrix. The rows and columns of the matrix correspond to the nodes in the initial node set.

[0181] Use the hierarchical clustering algorithm to perform cluster analysis on the node similarity matrix. Set the number of clusters to 4. According to the similarity values ​​in the node similarity matrix, cluster the nodes with high similarity into the same cluster. Store the clustering results in a list, where each element of the list represents a cluster, and the cluster contains the nodes contained in the cluster.

[0182] According to the node clustering results, a node hierarchy is constructed. Each cluster is regarded as a parent node, and the nodes in the cluster are child nodes. If a cluster contains only one node, the node is directly used as the parent node. For example, if the clustering result is [[age, hypertension], [diabetes], [average IPI, IPI change trend], [probability of disease worsening, nursing level requirement]], the constructed hierarchy is: four parent nodes represent four clusters respectively, the first two parent nodes have two child nodes respectively, and the last two parent nodes also have two child nodes respectively.

[0183] Use constraint-based structural learning algorithms, such as the PC algorithm, to learn the dependencies between nodes in the node hierarchy. Based on the preprocessed data, determine whether there is a dependency between the parent node and the child node, and whether there is a dependency between the child nodes. For example, determine whether age affects high blood pressure based on the data. Add the learned dependencies to the node hierarchy to obtain a node hierarchy with dependencies.

[0184] Based on the node hierarchy with dependencies, the `pgmpy` library is used to build a Bayesian network. The dependencies between parent nodes and child nodes are converted into directed edges in the Bayesian network. The resulting Bayesian network is stored in the `pgmpy` BayesianModel object format.

[0185] Preferably, step S4 comprises the following steps:

[0186] Step S41: performing individualized patient information retrieval according to the disease trend prediction vector to obtain a patient feature file;

[0187] Step S42: Acquire the nursing resource status according to the warning level data to obtain a list of available resources;

[0188] Step S43: generating an initial nursing plan according to the warning level data, the patient characteristic file and the available resource list to obtain an initial nursing plan;

[0189] Step S44: The initial nursing plan is adjusted individually according to the patient's characteristic profile and the list of available resources to obtain a recommended nursing plan.

[0190] In an embodiment of the present invention, a prediction vector of disease trend (PTV) is obtained. According to the patient ID, the individualized information of the patient is retrieved from the electronic medical record system, including age, gender, allergy history, past medical history, current diagnosis, laboratory test results, imaging test results, and the most recent nursing record. The retrieved information is stored in a data structure in JSON format to form a patient feature file. For example, the patient feature file contains the following information: `{"age": 65, "gender": "male", "allergy history": "penicillin", "past medical history": "hypertension", "current diagnosis": "pneumonia", ...}`.

[0191] Get the warning level data. According to the warning level, get the currently available nursing resource status information from the hospital resource management system. For example, get the number of available beds, the number of medical staff on duty, the number of ventilators, the number of monitors, etc. for primary, secondary, and tertiary care. Store the obtained resource status information in a JSON format data structure to form a list of available resources. For example, the list of available resources contains the following information: `{"primary nursing beds": 10, "secondary nursing beds": 20, "tertiary nursing beds": 5, "medical staff on duty": 15, "ventilators": 3, "monitors": 10, ...}`.

[0192] According to the warning level data, patient characteristic profiles, and available resource lists, the matching initial care plan is retrieved from the predefined care plan template library. The care plan template library contains care plan templates for different diseases, different warning levels, and different resource allocation situations. For example, for pneumonia patients, the warning level is 2, and there are sufficient secondary care beds. The initial care plan template may include: monitoring vital signs every 4 hours, medical staff care every 6 hours, and chest X-ray examination every day. The retrieved care plan template is used as the initial care plan.

[0193] Based on the patient's profile and the list of available resources, medical staff make personalized adjustments to the initial care plan. For example, if the patient is allergic to penicillin, the penicillin drugs in the initial care plan are replaced with other antibiotics. If the current ventilator resources are insufficient, the respiratory support plan is adjusted according to the patient's respiratory condition and blood oxygen saturation and other indicators. The dosage and frequency of medication are adjusted according to the patient's age and medical history. The adjusted care plan is used as the recommended care plan and output in JSON format, for example: `{"Monitoring frequency": "Every 4 hours", "Care plan": "Medical staff care, once a day", "Medication plan": "Cefuroxime sodium, twice a day", ...}`.

[0194] Preferably, step S5 comprises the following steps:

[0195] Step S51: Recording nursing execution status according to the recommended nursing plan to obtain nursing execution record data;

[0196] Step S52: Tracking the changes of physiological indicators according to the recommended nursing plan to obtain the trajectory of physiological indicator changes;

[0197] Step S53: Perform a preliminary evaluation of nursing effects according to the nursing execution record data and the physiological index change trajectory to obtain preliminary effect evaluation data;

[0198] Step S54: dynamically adjusting the nursing plan according to the preliminary effect evaluation data and the recommended nursing plan to obtain an adjusted nursing plan;

[0199] Step S55: Generate nursing effect indicators based on the preliminary effect evaluation data and the adjusted nursing plan to obtain nursing effect indicators.

[0200] In an embodiment of the present invention, medical staff record the execution of the recommended nursing plan through a mobile nursing terminal or an electronic nursing record system. After each nursing operation is performed, the medical staff records the specific content of the operation, the execution time, the executor, and the patient's response in the system. For example, if the medical staff performs "medical staff care", the operation time, drug name, dosage, changes in the patient's respiratory condition and other information are recorded. The system automatically stores this information in the database to form nursing execution record data, and associates it with the patient ID and recommended nursing plan. The data is stored in JSON format, for example: `{"operation":"medical staff care","time":"2024-10-2710:00:00","executor":"medical staff Zhang San","drug":"salbutamol","dose":"2.5mg","patient response":"cough relieved",...}`

[0201] The system continuously collects the patient's physiological indicator data, such as IPI, heart rate, blood pressure, blood oxygen saturation, respiratory rate, etc. from IoT sensors and nursing monitoring equipment. According to the monitoring frequency specified in the recommended nursing plan, such as once an hour, the physiological indicator data before and after the nursing execution is recorded. These data are arranged in chronological order to form a physiological indicator change trajectory. For example, if the recommended nursing plan stipulates that IPI is monitored once an hour, the system records the IPI value once an hour and associates these values ​​with the corresponding nursing execution time to form an IPI change trajectory, which is stored in the form of time series data.

[0202] The system conducts a preliminary evaluation of the nursing effect based on the nursing execution record data and the change trajectory of physiological indicators. For example, compare the changes in IPI values ​​before and after nursing execution to determine whether the nursing effect is positive, negative, or no significant change. If the IPI value rises, the nursing effect is considered positive; if the IPI value decreases, the nursing effect is considered negative; if the IPI value does not change much, the nursing effect is considered to have no significant change. Quantify the evaluation results, for example, use 1 for positive, -1 for negative, and 0 for no significant change. Use the quantified evaluation results as preliminary effect evaluation data

[0203] The system dynamically adjusts the recommended care plan based on the preliminary effect evaluation data. If the preliminary effect evaluation data shows that the care effect is negative, the system will adjust the care plan based on pre-set rules or in combination with the expert knowledge base. For example, if the patient's IPI value continues to decrease, the system may recommend increasing the monitoring frequency, adjusting the medication dosage, or changing the care plan. The adjusted care plan will be used as the new recommended care plan. If the preliminary effect evaluation data shows that the care effect is positive or there is no significant change, the original recommended care plan will remain unchanged.

[0204] The system generates a nursing effectiveness index (CEI) based on the preliminary effect evaluation data and the adjusted nursing plan. CEI can be a comprehensive indicator, such as the weighted sum of indicators such as IPI change, patient comfort score, and nursing cost. It can also be a combination of multiple independent indicators, such as IPI change rate, complication rate, length of stay, etc. CEI is used to quantify nursing effects and provide data support for subsequent nursing quality evaluation and improvement. For example, CEI can be defined as: `CEI = 0.5*IPI change rate + 0.3*patient comfort score - 0.2*nursing cost`.

[0205] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0206] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A digital management method for graded nursing care in intensive care units based on the Internet of Things, characterized in that: The following steps are involved: Step S1: real-time collection of heterogeneous data of patients through nursing monitoring equipment to obtain principle physiological data stream; Calculate comprehensive physiological indexes on the principle physiological data stream to obtain comprehensive physiological indexes; Step S2: Predict individualized physiological indicators based on the comprehensive physiological indicators, and predict the probability of the patient's condition worsening to obtain condition worsening probability data; predict the nursing level demand based on the prediction results of the individualized physiological indicators to obtain nursing level demand probability data; generate a condition trend prediction vector for the condition worsening probability data and the nursing level demand probability data to obtain a condition trend prediction vector; Step S3: setting a dynamic warning threshold according to the disease trend prediction vector and the preset warning rules to obtain a dynamic threshold matrix; triggering a warning signal according to the disease trend prediction vector and the dynamic threshold matrix, and determining the warning level to obtain warning level data; Step S4: recommending nursing resources based on the disease trend prediction vector and the warning level data to obtain a recommended nursing plan; Step S5: recommend a nursing plan to track changes in physiological indicators and obtain a trajectory of changes in physiological indicators; perform nursing effect evaluation based on the trajectory of changes in physiological indicators and obtain nursing effect indicators to achieve the digital management task of graded nursing in the intensive care unit.

2. The method for digital management of graded nursing care in intensive care units based on the Internet of Things according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: real-time collection of heterogeneous data of the patient through nursing monitoring equipment to obtain a principle physiological data stream; Step S12: Associating the original physiological data stream with the patient identity to obtain patient identity associated data; performing physiological data synchronization processing according to the patient identity associated data to obtain patient synchronized physiological data; Step S13: extracting physiological data features from the patient's synchronous physiological data to obtain a physiological feature vector; Step S14: Calculate the comprehensive physiological index according to the physiological characteristic vector to obtain the comprehensive physiological index.

3. The method for digital management of graded nursing care in intensive care units based on the Internet of Things according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: constructing a time series of the comprehensive physiological index to obtain a time series of the comprehensive physiological index; Step S22: performing individualized physiological pattern learning on the comprehensive physiological index time series to obtain an individualized physiological model; Step S23: using the individualized physiological model to predict individualized physiological indicators to obtain individualized physiological indicator prediction results; Step S24: predicting the probability of worsening of the patient's condition based on the prediction results of the individualized physiological indicators to obtain data on the probability of worsening of the condition; Step S25: Predicting nursing level requirements based on individualized physiological index prediction results and preset nursing level classification rules to obtain nursing level requirement probability data; Step S26: Generate a disease trend prediction vector based on the disease deterioration probability data and the nursing level requirement probability data to obtain a disease trend prediction vector.

4. The method for digital management of graded nursing care in intensive care units based on the Internet of Things according to claim 3 is characterized in that: Step S25 includes the following steps: Step S251: Predicting the change trend of physiological indicators based on the individualized physiological indicator prediction results to obtain a nursing demand trend prediction sequence; Step S252: performing dynamic and static nursing level matching according to the nursing demand trend prediction sequence and the preset nursing level division rules to obtain a nursing demand matching result, wherein the preset nursing level division rules include static division rules and dynamic division rules; Step S253: constructing a nursing demand probability model according to the nursing demand matching result to obtain a nursing demand probability model; calculating the nursing demand probability according to the nursing demand probability model and the nursing demand matching result to obtain a nursing demand probability matrix; Step S254: Obtain nursing resource configuration data; optimize the nursing resource demand probability matrix according to the nursing resource configuration data to obtain nursing level demand probability data.

5. The method for digital management of graded nursing care in intensive care units based on the Internet of Things according to claim 4 is characterized in that: Step S252 is specifically as follows: Perform point-by-point matching on the nursing demand trend prediction sequence and the static division rule to obtain point-by-point matching results; perform preliminary marking based on the point-by-point matching results to obtain static matching results; Extract trend features from the nursing demand trend prediction sequence to obtain trend features; Dynamic rule matching is performed on trend features and dynamic division rules to obtain dynamic matching results; The static matching results and the dynamic matching results are fused, and matching conflicts are eliminated to obtain fused matching results. The nursing needs matching result is generated according to the fusion matching result to obtain the nursing needs matching result.

6. The method for digital management of graded nursing care in intensive care units based on the Internet of Things according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: setting a dynamic warning threshold according to the disease trend prediction vector and the preset warning rules to obtain a dynamic threshold matrix; Step S32: performing single-index avoidance comparison on the disease trend prediction vector and the dynamic threshold matrix to obtain a single-index comparison result; performing multi-index fusion on the single-index comparison result to obtain a multi-index fusion result; Step S33: Perform early warning trigger judgment according to the multi-indicator fusion result and the dynamic threshold matrix to obtain an early warning trigger signal; Step S34: Calculate the risk factor weights according to the disease trend prediction vector, the patient's historical physiological data, and the preset warning classification results to obtain a risk factor weight table; Step S35: Calculate the risk score according to the risk factor weight table and the early warning trigger signal to obtain a comprehensive risk score; Step S36: Classify the warning level according to the comprehensive risk score and the preset warning classification results to obtain warning level data.

7. The method for digital management of graded nursing care in intensive care units based on the Internet of Things according to claim 6 is characterized in that: Step S31 includes the following steps: Step S311: Acquire the patient's historical physiological data; construct an early warning Bayesian network based on the patient's historical physiological data and the disease trend prediction vector to obtain an early warning Bayesian network; Step S312: using the patient's historical physiological data to set the probability distribution of the early warning Bayesian network, and obtaining a probabilistic Bayesian network; Step S313: extracting the patient's current physiological index from the principle physiological data stream to obtain the patient's current physiological index; Step S314: input the patient's current physiological indicators and the disease trend prediction vector into the probabilistic Bayesian network to perform dynamic threshold calculation to obtain a dynamic threshold matrix.

8. The method for digital management of graded nursing care in intensive care units based on the Internet of Things according to claim 7 is characterized in that: Step S311 is specifically as follows: Perform data preprocessing on the patient's historical physiological data and disease trend prediction vector to obtain preprocessed data; Define the initial node set according to the preprocessed data to obtain the initial node set; Calculate node similarity based on preprocessed data and initial node set to obtain node similarity matrix; Perform node clustering analysis on the node similarity matrix to obtain node clustering results; construct a hierarchical structure based on the node clustering results and the initial node set to obtain a node hierarchy; Using the preprocessed data to learn the dependency relationship of the node hierarchy, a node hierarchy with dependency relationship is obtained; A Bayesian network is constructed according to the node hierarchy with dependency relationships to obtain an early warning Bayesian network.

9. The method for digital management of graded nursing care in intensive care units based on the Internet of Things according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: performing individualized patient information retrieval according to the disease trend prediction vector to obtain a patient feature file; Step S42: Acquire the nursing resource status according to the warning level data to obtain a list of available resources; Step S43: generating an initial nursing plan according to the warning level data, the patient characteristic file and the available resource list to obtain an initial nursing plan; Step S44: The initial nursing plan is adjusted individually according to the patient's characteristic profile and the list of available resources to obtain a recommended nursing plan.

10. The method for digital management of graded nursing care in intensive care units based on the Internet of Things according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: Recording nursing execution status according to the recommended nursing plan to obtain nursing execution record data; Step S52: Tracking the changes of physiological indicators according to the recommended nursing plan to obtain the trajectory of physiological indicator changes; Step S53: Perform a preliminary evaluation of nursing effects according to the nursing execution record data and the physiological index change trajectory to obtain preliminary effect evaluation data; Step S54: dynamically adjusting the nursing plan according to the preliminary effect evaluation data and the recommended nursing plan to obtain an adjusted nursing plan; Step S55: Generate nursing effect indicators based on the preliminary effect evaluation data and the adjusted nursing plan to obtain nursing effect indicators.

Images