Clinical closed-loop nursing management method based on hospital chronic disease nursing big data

Abstract

The invention provides a clinical closed-loop nursing management method based on hospital chronic disease nursing big data, and the method comprises the steps: constructing a closed-loop nursing management process of data collection-risk assessment-intervention generation-execution tracking-effect evaluation-loop optimization, and fusing the algorithms of time sequence prediction, real-time anomaly detection, causal inference, federated learning and the like. The chronic disease nursing data of the patient is comprehensively collected and preprocessed, so that the integrity and accuracy of the data are ensured. And by using the pre-trained LSTM risk prediction model, potential information in the data can be fully mined, and the chronic disease deterioration risk level of the patient can be accurately output. A propensity score matching (PSM) model is employed to evaluate the effectiveness of the intervention plan. The model can effectively control the influence of confounding factors and accurately evaluate the actual effect of the intervention plan by calculating the tendency score of the intervention accepted by the patient. Therefore, personalized, precise and dynamic chronic disease nursing management is realized.

Description

Technical Field

[0001] This invention relates to the field of clinical nursing technology, and in particular to a clinical closed-loop nursing management method based on hospital chronic disease nursing big data, as well as a clinical closed-loop nursing management device, electronic device, and computer-readable storage medium based on hospital chronic disease nursing big data. Background Technology

[0002] Traditional chronic disease care lacks a comprehensive and systematic data collection and preprocessing mechanism. In most cases, it relies solely on data from patients' regular hospital checkups, resulting in a single and discontinuous data source that fails to reflect the patient's health status in real time. For example, it's difficult to comprehensively collect data on patients' daily exercise, diet, and other lifestyle habits, leading to an inability to comprehensively analyze the patient's overall health. Even when some data is collected, it is not fully utilized for risk assessment. The lack of advanced LSTM risk prediction models like those used in this process prevents accurate quantification and classification of the risk of chronic disease exacerbation, resulting in a lack of targeted care measures.

[0003] Furthermore, current nursing care often employs generic protocols without fully considering individual patient characteristics such as age, gender, complications, and medication history. Different patients have different physical conditions and disease progression, and generic protocols cannot meet the actual needs of each patient. For example, elderly and younger patients should have different focuses in their diet and exercise plans, but traditional nursing care cannot be individualized. It also fails to incorporate historical intervention data to develop intervention plans, making it impossible to adjust and optimize based on previous nursing outcomes, thus compromising the effectiveness of the nursing plan.

[0004] Traditional chronic disease care lacks real-time monitoring methods for patient health data, typically relying on regular hospital checkups, which fails to detect abnormal changes in patient health status promptly. When a patient's condition worsens, timely intervention is often delayed, hindering treatment. The absence of an isolated forest anomaly detection model, as used in this process, for real-time health data analysis makes it difficult to quickly and accurately identify abnormalities and issue timely alerts, preventing caregivers from promptly understanding patients' emergency situations. Current nursing care lacks scientific methods and models for evaluating the effectiveness of intervention programs. Evaluations are often based solely on subjective patient feedback or limited indicators, lacking an objective and comprehensive assessment system. The absence of a propensity score matching (PSM) model for effectiveness evaluation fails to effectively control for confounding factors, making it difficult to accurately compare the effects between intervention-received and non-intervention-received patients and determine the actual effectiveness of the intervention program. Summary of the Invention

[0005] To address the technical problems existing in the prior art, the present invention provides the following technical solution: On the one hand, a closed-loop clinical nursing management method based on big data of chronic disease care in hospitals is provided. This method is implemented by electronic devices and includes: S1. Data Collection: Collect and preprocess patients' chronic disease care data; S2. Risk Assessment: Input the pre-processed chronic disease care data into the pre-trained LSTM risk prediction model to output the patient's risk level of chronic disease exacerbation; S3. Intervention Plan Generation: Based on the risk level, combined with the patient's individual characteristics (age, gender, complications, medication history) and historical intervention effect data, a personalized nursing intervention plan is generated through a rule engine and collaborative filtering recommendation algorithm. The nursing intervention plan includes dietary guidance, exercise plan, medication reminders, and follow-up arrangements. S4. Execution Tracking: Collect patients' health data in real time through wearable devices, input the real-time health data into a pre-trained isolated forest anomaly detection model, and output anomaly scores; when the anomaly scores exceed a preset threshold, trigger an alarm and send the anomaly information to nursing staff; S5. Effect Evaluation: Collect effect data of patients after receiving the intervention, and evaluate the effect of the intervention plan using a propensity score matching (PSM) model; wherein, the processing procedure of the PSM model includes: calculating the propensity score of patients to receive the intervention, matching patients who received the intervention with patients who did not receive the intervention according to their propensity scores, and comparing the effect indicators of the two groups; S6. Iterative Optimization: Feedback the effect evaluation results to the LSTM risk prediction model in S2 and the intervention plan generation module in S3 to update the model parameters and rules.

[0006] Preferably, the chronic disease care data includes electronic medical record data, nursing record data, laboratory test data, wearable device data, and patient self-report data.

[0007] Preferably, the preprocessing includes missing value processing, outlier processing, data standardization, and time-series data alignment.

[0008] Preferably, the training process of the LSTM risk prediction model includes: acquiring historical chronic disease care time series data and corresponding deterioration event labels of patients, inputting the time series data into the LSTM network, updating the cell state and hidden state through the forget gate, input gate, and output gate, outputting the risk prediction value, and optimizing the model parameters using the cross-entropy loss function; The forget gate formula for the LSTM network is:

[0009] in, It is the sigmoid activation function. Here is the forget gate weight matrix. This is the hidden state from the previous moment. Input data for the current moment. (For forget gate bias term) The input gate formula is:

[0010] in, The input gate weight matrix, For input gate bias terms; tanh is the hyperbolic tangent activation function. This is the candidate cell state weight matrix. This refers to the candidate cell state bias term; The cell state update formula is: , in, This represents the element-wise product (Hadamard product). The output gate formula is , This is the output gate weight matrix. (for output gate bias term) ( (Hidden state at the current moment).

[0011] Preferably, the anomaly score formula of the isolated forest anomaly detection model is: , Where s(x,n) is the anomaly score of sample x, n is the total number of samples, E(h(x)) is the average path length of sample x in all isolated trees in the isolated forest, and c(n) is the average path length of n samples (normalization factor).

[0012] Preferably, the LSTM risk prediction model in S2 is trained using a federated learning algorithm, specifically the FedAvg algorithm, which includes the following steps: S21. The server initializes global model parameters; S22. Each client (hospital) downloads the global model parameters, trains the model using local chronic disease care data, and obtains the local model parameters; S23. Each client sends its local model parameters to the server; S24. The server averages the local model parameters of all clients to obtain new global model parameters; S25. Repeat steps S22-S24 until the model converges.

[0013] On the other hand, a clinical closed-loop nursing management device based on hospital chronic disease nursing big data is provided. This device is used to implement the aforementioned clinical closed-loop nursing management method based on hospital chronic disease nursing big data. The device includes: Data acquisition module: Used to collect patients' chronic disease care data and perform preprocessing; Risk assessment module: used to input pre-processed chronic disease care data into a pre-trained LSTM risk prediction model and output the patient's risk level of chronic disease exacerbation; Intervention plan generation module: used to generate personalized nursing intervention plans based on the risk level, combined with the patient's individual characteristics and historical intervention effect data; Execution tracking module: used to collect patients' health data in real time through wearable devices, input it into the isolated forest anomaly detection model, output anomaly scores and trigger alarms; The effect evaluation module is used to collect effect data on patients after receiving intervention and to evaluate the effectiveness of the intervention plan using the PSM model. The iterative optimization module is used to feed back the effect evaluation results to the risk assessment module and the intervention plan generation module, and update the model parameters and rules.

[0014] On the other hand, an electronic device is provided, comprising: a processor; and a memory storing computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, any one of the methods described above in the clinical closed-loop nursing management method based on hospital chronic disease nursing big data is implemented.

[0015] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement any of the above-described methods of clinical closed-loop nursing management based on big data of chronic disease care in hospitals.

[0016] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: This invention ensures the integrity and accuracy of patients' chronic disease care data by comprehensively collecting and preprocessing the data. Using a pre-trained LSTM risk prediction model, the potential information within the data is fully extracted, accurately outputting the patient's risk level for chronic disease exacerbation. This risk assessment method based on big data and deep learning is more scientific and accurate than traditional experience-based judgment, providing a strong basis for subsequent nursing interventions. Simultaneously, by combining individual patient characteristics (age, gender, complications, medication history) and historical intervention effect data, a personalized nursing intervention plan is generated using a rule engine and collaborative filtering recommendation algorithm. This plan includes dietary guidance, exercise plans, medication reminders, follow-up arrangements, and other aspects, meeting the specific needs of different patients and improving the targeting and effectiveness of nursing care.

[0017] Propensity score matching (PSM) was used to evaluate the effectiveness of the intervention plan. This model calculates patients' propensity scores for receiving the intervention, matches those receiving the intervention with those not receiving it based on their propensity scores, and compares the effectiveness indicators between the two groups. This effectively controls for the influence of confounding factors and accurately assesses the actual effect of the intervention plan. This scientific evaluation method provides a reliable basis for optimizing nursing care plans. The effectiveness evaluation results are fed back to the LSTM risk prediction model and the intervention plan generation module to update model parameters and rules. Through continuous learning and optimization, risk assessment becomes more accurate, and intervention plans become more rational, forming a closed-loop optimization system. With the continuous accumulation of patient data and continuous model updates, nursing care outcomes will be continuously improved. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a clinical closed-loop nursing management method based on big data of chronic disease care in hospitals, provided by an embodiment of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0021] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0022] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0023] In this embodiment of the invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0024] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0025] This invention provides a clinical closed-loop nursing management method based on big data in hospital chronic disease care. This method can be implemented using electronic devices, such as terminals or servers. Figure 1 The flowchart shown is a clinical closed-loop nursing management method based on big data of chronic disease care in hospitals. The processing flow of this method may include the following steps: S1. Data Collection: Collect and preprocess patients' chronic disease care data; S2. Risk Assessment: Input the pre-processed chronic disease care data into the pre-trained LSTM risk prediction model to output the patient's risk level of chronic disease exacerbation; S3. Intervention Plan Generation: Based on the risk level, combined with the patient's individual characteristics (age, gender, complications, medication history) and historical intervention effect data, a personalized nursing intervention plan is generated through a rule engine and collaborative filtering recommendation algorithm. The nursing intervention plan includes dietary guidance, exercise plan, medication reminders, and follow-up arrangements. S4. Execution Tracking: Collect patients' health data in real time through wearable devices, input the real-time health data into a pre-trained isolated forest anomaly detection model, and output anomaly scores; when the anomaly scores exceed a preset threshold, trigger an alarm and send the anomaly information to nursing staff; S5. Effect Evaluation: Collect effect data of patients after receiving the intervention, and evaluate the effect of the intervention plan using a propensity score matching (PSM) model; wherein, the processing procedure of the PSM model includes: calculating the propensity score of patients to receive the intervention, matching patients who received the intervention with patients who did not receive the intervention according to their propensity scores, and comparing the effect indicators of the two groups; S6. Iterative Optimization: Feedback the effect evaluation results to the LSTM risk prediction model in S2 and the intervention plan generation module in S3 to update the model parameters and rules.

[0026] The working principle of this embodiment will be described in detail below.

[0027] This solution constructs a closed-loop nursing management process encompassing "data collection - risk assessment - intervention generation - execution tracking - effect evaluation - iterative optimization," integrating algorithms such as time-series prediction, real-time anomaly detection, causal inference, and federated learning to achieve personalized, precise, and dynamic chronic disease nursing management. The specific steps are as follows: 1. Data Acquisition: Multi-source data fusion and preprocessing Data source: Collection of patients' chronic disease care data throughout the entire lifecycle, including: Structured data: Diagnosis, medication, laboratory tests (blood glucose, blood pressure, blood lipids, etc.) and nursing records (intake, output, pressure ulcer scores, etc.) in electronic medical records (EHR); Unstructured data: Patient self-reports (symptoms, diet, exercise), nursing documentation (medical order execution records); Real-time data: Time-series data such as blood glucose, blood pressure, heart rate, and steps collected by wearable devices (smartwatches, blood glucose meters).

[0028] Preprocessing steps: Missing value handling: For time series data (such as continuous blood glucose monitoring), linear interpolation is used to fill short-term missing values ​​(<6 hours), and spline interpolation is used to fill long-term missing values ​​(6-24 hours); in structured data, categorical variables (such as complication types) are filled with the mode, and continuous variables (such as blood pressure) are filled with the median (to avoid the influence of extreme values). Outlier handling: The Winsorization method (truncating the upper and lower 1% extreme values) is used to correct abnormal data from wearable devices, where the threshold is dynamically adjusted using the IQR rule (interquartile range ± 1.5 × IQR); obvious physiological abnormal values ​​(such as blood glucose < 2.8 mmol / L or > 33.3 mmol / L) are directly marked as abnormal and trigger manual review. Standardization: Z-score standardization is adopted. ,in The mean, To standardize the data scale (standard deviation), non-normally distributed features (such as heart rate) are subjected to Box-Cox transformation before standardization. Time alignment: Wearable device data is aligned hourly by time stamp calibration, and Kalman filtering is used to fuse data from multiple devices (such as smartwatches and blood glucose meters) to form a 24-dimensional time-series feature vector for each patient's day.

[0029] 2. Risk Assessment: LSTM-based Prediction of Chronic Disease Exacerbation Risk Algorithm selection: Long Short-Term Memory (LSTM) network is used to process chronic disease time series data (such as blood glucose changes over 7 consecutive days). Bidirectional LSTM is used to capture long-term dependencies, and attention mechanism is combined to strengthen the weight of key time points (such as postprandial blood glucose peak) to predict the patient's risk level of chronic disease exacerbation (low, medium, high) in the next 1-3 months.

[0030] LSTM model structure and formulas: The core of LSTM is cell state ( And three gating mechanisms (forget gate, input gate, output gate), the formulas are as follows: Forgotten Gate ( ): Decide to discard the cell state from the previous moment ( The information is output in the range [0,1] (1 indicates that it is completely kept, and 0 indicates that it is completely discarded):

[0031] in, It is the sigmoid activation function. Here is the forget gate weight matrix. The output represents the hidden state from the previous moment. Input the current time (e.g., the patient's blood glucose and blood pressure data at time t). For bias.

[0032] Input gate ( , Update cell state: Input gate activation ( ): Decide how much new information to retain:

[0033] Candidate cell status ( New candidate information (range [-1, 1]) is generated using the tanh activation function.

[0034] Cell state update: The current cell state is updated by fusing the results of the forget gate and the input gate.

[0035] in, It is an element-wise product (Hadamard product).

[0036] Output gate ( ): Determines how much cell state information to output to the hidden state:

[0037]

[0038] in, The hidden state at the current moment (model output). Compress the cell state to [-1, 1].

[0039] Model training and application: Training data: historical patients' time-series nursing data (time window = 7 days, sampling interval = 1 hour) + exacerbation event labels (such as whether complications occurred, hospitalization), using 5-fold cross-validation to avoid overfitting; Loss function: Cross-entropy loss is used. ,in For real labels, To predict probabilities, Focal Loss is combined to solve the sample imbalance problem (the weight of high-risk samples is increased by 2 times). Application: Patient data is acquired in real time through the hospital's HIS system interface, input into a trained LSTM model (deployed as a TensorFlow Serving service), outputs the probability of chronic disease exacerbation (e.g., 85%), maps it to a risk level (low risk <30%, medium risk 30%-70%, high risk ≥70%), and generates a risk report.

[0040] 3. Intervention plan generation: Personalized recommendations using a rule engine and collaborative filtering Core logic: Based on the patient's risk level (e.g., high risk), individual characteristics (age, gender, complications, medication history, dietary preferences), and historical intervention effects (e.g., the blood glucose control rate of a certain type of dietary guidance for similar patients), a personalized nursing intervention plan is generated.

[0041] Rule Engine: The rule base is built based on the "Guidelines for the Prevention and Treatment of Type 2 Diabetes in China (2024 Edition)" and contains 200+ clinical rules (such as "high-risk diabetic patients need to monitor blood glucose 4 times a day" and "metformin is contraindicated in patients with renal insufficiency"), which are parsed and executed in real time through the Drools rule engine; Collaborative filtering recommendation: User-based collaborative filtering (CF) is used to calculate patient similarity using weighted cosine similarity (complication weight 0.4, medication history weight 0.3, age weight 0.2, gender weight 0.1). Effective intervention measures for the top-3 similar patients are recommended (e.g., "Patient A and Patient B have a similarity of 0.8. Patient B's blood sugar decreased by 20% after adopting a 'low-carb diet + 30 minutes of walking every day'. Patient A is recommended to adopt this plan").

[0042] 4. Execution Tracking: Real-time Anomaly Monitoring Based on Isolation Forest Algorithm selection: Isolation Forest is used to process real-time data from wearable devices (sampling frequency = 5 minutes / time). Anomaly detection is achieved by constructing 100 isolated trees (sample size = 256 per tree), with an average detection latency of <10 seconds.

[0043] The formula for outlier scoring in isolated forests: For each sample x, the anomaly score s(x,n) is defined as: , in: E(h(x)): The average path length of sample x across all isolated trees in the isolated forest (the shorter the path, the more likely it is to be an anomaly); c(n): Correction term. When n>2, c(n)=2H(n-1)-2(n-1) / n (H is the harmonic function); n: number of training samples; s(x,n)∈[0,1], s(x,n→1 represents Propensity score (PS) calculation: using a logistic regression model ( ,in This indicates acceptance of intervention; X represents patient characteristics. It is the sigmoid activation function. For the intercept term, ,..., Features ,..., (corresponding regression coefficients) to calculate the probability of each patient receiving intervention; The range of s(x,n) is [0,1] (s→1 indicates an anomaly, s→0 indicates normal).

[0044] Application process: Training: Train the isolated forest model using real-time data from historical patients (build 100 isolated trees); Real-time monitoring: Input the patient's real-time blood glucose and blood pressure data into the model to calculate anomaly scores; Alarm Trigger: When the abnormal score is ≥0.9 (preset threshold), the system sends an SMS / APP alarm to the nursing staff (e.g., "Patient Zhang San, blood glucose 16.2 mmol / L, abnormal score 0.95, please handle immediately").

[0045] 5. Effect Evaluation: Causal Inference of Intervention Effects Based on PSM Algorithm selection: Propensity score matching (PSM) was used to evaluate the effectiveness of the intervention plan, excluding the influence of confounding factors (such as the patient's baseline health status and age) to obtain the net effect of the intervention (e.g., "patients who adopted this intervention plan had a 30% higher glycemic control rate than those who did not adopt it").

[0046] The core steps of PSM: Propensity score (PS) calculation: using a logistic regression model ( ,in (Indicates acceptance of intervention, X represents patient characteristics). Key features (such as age, glycated hemoglobin, and number of complications) are screened using L1 regularization (Lasso) to avoid overfitting. Matching: Nearest neighbor matching (1:1 matching) was used, with the caliper range set at 0.02 times the standard deviation of the propensity score. Each patient receiving intervention was matched with one non-intervention patient whose propensity score was closest to the intervention score. After matching, the balance was tested by the standardized mean difference (SMD) test to ensure that the SMD of all covariates was <0.1. Effectiveness assessment: After matching, the efficacy indicators of the two groups of patients were compared, including glycemic control rate (number of days with glycemic control after intervention / total number of days × 100%), complication rate (number of new complications / total number of cases × 100%), and patient satisfaction (5-point Likert scale score). The mean treatment effect (ATE) and 95% confidence interval were calculated using t-test or chi-square test.

[0047] 6. Iterative Optimization: Feedback-Based Model Iteration Core logic: Feedback of effectiveness evaluation results (e.g., "A certain intervention plan only achieved a 50% blood glucose control rate in high-risk patients") to the risk assessment module and intervention generation module to optimize model parameters. Risk assessment model optimization: The LSTM model is retrained using an incremental training strategy (50% new data + 50% historical data), with a learning rate decay set (initially 0.001, decaying by 10% every 5 rounds). Overfitting is avoided by using an early stopping method (stopping if the validation set loss does not decrease after 3 rounds). The model performance index should reach AUC≥0.85. Intervention generation module optimization: Establish a rule base version management mechanism (e.g., update from V1.0 to V1.1), new rules must be verified through A / B testing (the experimental group uses the new rules, the control group uses the old rules, and the sample size is 100 cases in each group), and the new rules will be officially launched when the blood glucose control rate improves by ≥15%; increase the weight of the "dietary preference" feature in collaborative filtering recommendation (from 0.1 to 0.2), and use weighted cosine similarity for similarity calculation.

[0048] This solution can also use the FedAvg algorithm to achieve cross-hospital model collaborative training, improving the model's generalization ability without sharing the original data.

[0049] The core formula of FedAvg: Global model parameters Based on the local model parameters of each hospital (client) On average: , Where K is the number of hospitals participating in federal learning, and t is the number of training rounds.

[0050] The application process is as follows: Initialization: The server (such as the cloud platform of a hospital group) initializes the global LSTM model parameters; Local training: Each hospital trains the model using local chronic disease data (≥1000 patients), employing the Adam optimizer (learning rate 0.001, β1=0.9, β2=0.999), batch size 32, and training for 10 epochs; differential privacy technology (adding Laplace noise, ε=1.0) is used to protect data privacy and ensure that the local model parameters do not leak the original data; Parameter upload: Each hospital sends its local model parameters to the server; Global update: The server aggregates parameters by weighted average based on the data volume of each hospital (data volume ratio = hospital sample size / total sample size, weight = data volume ratio). If the parameters of a hospital are abnormal (deviation from the global mean > 3 times the standard deviation), the update round is removed to ensure the stability of the global model. Iterative loop: Repeat local training - parameter upload - global update until the model converges (e.g., the loss function drops below 0.1).

[0051] Example: Closed-loop nursing management case of patients with type 2 diabetes Patient basic information: Patient Wang, male, 65 years old, with a 10-year history of type 2 diabetes and hypertension (150 / 90 mmHg), medication history: metformin (0.5g / dose, twice daily), dietary preference: high-carbohydrate diet.

[0052] (1) Data acquisition and preprocessing: Structured data: Fasting blood glucose (mean 8.5 mmol / L, standard deviation 1.2) and glycated hemoglobin (HbA1c) of 7.8% over the past 3 months in the EHR; Unstructured data: Patient self-reported "200g of staple food per day, insufficient exercise"; Real-time data: The smart blood glucose meter collects data hourly, with 6 hours missing for 3 days (filled using spline interpolation). Two outliers were detected (16.2 mmol / L and 2.5 mmol / L, corrected to 12.0 mmol / L and 3.9 mmol / L after Winsorization). After time alignment, a daily 24-dimensional blood glucose sequence is generated (e.g., 6.8 mmol / L at 6:00 and 9.2 mmol / L at 12:00).

[0053] (2) Risk assessment: Input LSTM model (bidirectional LSTM + attention mechanism), output the probability of deterioration in the next 3 months is 82% (high risk), and the model feature importance shows that "2 hours postprandial blood glucose" has the highest weight (0.32).

[0054] (3) Intervention plan generation: Rule engine: High-risk patients should monitor their blood glucose 5 times a day (3 times before meals + 2 times after meals); Collaborative filtering: Matching similar patients (cosine similarity 0.85, same age, and comorbid hypertension), recommending an intervention plan of "low-carb diet (100g of staple food / day) + 40 minutes of brisk walking every day".

[0055] (4) Follow-up: On the 14th day of intervention, the wearable device monitored the post-lunch blood glucose in real time and found it to be 15.8 mmol / L. The isolated forest model abnormal score was 0.93 (≥0.9 threshold). The system triggered an alarm: "Patient Wang, blood glucose 15.8 mmol / L, abnormal score 0.93, please handle immediately." The nursing staff guided the adjustment of insulin dosage by telephone.

[0056] (5) Evaluation of efficacy: PSM-matched 1:1 control patients (propensity score 0.72 vs 0.73) were used. The blood glucose control rate of the intervention group (Wang) was 78% after 3 months, while that of the control group was 52%, and ATE was 26% (95% CI: 18%-34%, P<0.05).

[0057] (6) Cyclic optimization: Because the blood glucose control rate of high-risk patients was 78%, which did not meet the target (≥85%), the LSTM model was updated (the "insulin dose" feature was added), and the rule engine adjusted the monitoring frequency of high-risk patients to 6 times / day. After 3 months, the blood glucose control rate of similar patients increased to 87%.

[0058] Therefore, this solution integrates algorithms such as LSTM, isolated forest, PSM, and federated learning to realize an innovative process of "closed-loop nursing + personalized intervention + real-time monitoring". It solves the technical problems of "inaccurate risk prediction, insufficient personalized intervention, and unrigorous effect evaluation" in existing chronic disease nursing. It can be directly applied to the chronic disease nursing management system of hospitals to improve nursing efficiency and patient outcomes (such as blood glucose control rate and complication rate), and has industrial application value.

[0059] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0060] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more detailed understanding.

[0061] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0062] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0063] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0064] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0065] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0066] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0067] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0068] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0069] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A clinical closed-loop nursing management method based on big data of chronic disease care in hospitals, characterized in that, The method includes: S1. Data Collection: Collect and preprocess patients' chronic disease care data; S2. Risk Assessment: Input the pre-processed chronic disease care data into the pre-trained LSTM risk prediction model to output the patient's risk level of chronic disease exacerbation; S3. Intervention Plan Generation: Based on the risk level, combined with the patient's individual characteristics (age, gender, complications, medication history) and historical intervention effect data, a personalized nursing intervention plan is generated through a rule engine and collaborative filtering recommendation algorithm. The nursing intervention plan includes dietary guidance, exercise plan, medication reminders, and follow-up arrangements. S4. Execution Tracking: Collect patients' health data in real time through wearable devices, input the real-time health data into a pre-trained isolated forest anomaly detection model, and output anomaly scores; when the anomaly scores exceed a preset threshold, trigger an alarm and send the anomaly information to nursing staff; S5. Effect Evaluation: Collect effect data of patients after receiving the intervention, and evaluate the effect of the intervention plan using a propensity score matching (PSM) model; wherein, the processing procedure of the PSM model includes: calculating the propensity score of patients to receive the intervention, matching patients who received the intervention with patients who did not receive the intervention according to their propensity scores, and comparing the effect indicators of the two groups; S6. Iterative Optimization: Feedback the effect evaluation results to the LSTM risk prediction model in S2 and the intervention plan generation module in S3 to update the model parameters and rules.

2. The clinical closed-loop nursing management method based on hospital chronic disease nursing big data as described in claim 1, characterized in that, The chronic disease care data includes electronic medical record data, nursing record data, laboratory test data, wearable device data, and patient self-report data.

3. The clinical closed-loop nursing management method based on hospital chronic disease nursing big data as described in claim 1, characterized in that, The preprocessing includes missing value handling, outlier handling, data standardization, and time-series data alignment.

4. The clinical closed-loop nursing management method based on hospital chronic disease nursing big data as described in claim 1, characterized in that, The training process of the LSTM risk prediction model includes: acquiring historical chronic disease care time series data and corresponding deterioration event labels of patients; inputting the time series data into the LSTM network; updating cell state and hidden state through forget gate, input gate, and output gate; outputting risk prediction value; and optimizing model parameters using cross-entropy loss function. The forget gate formula for the LSTM network is: in, It is the sigmoid activation function. Here is the forget gate weight matrix. This is the hidden state from the previous moment. Input data for the current moment. (For forget gate bias term) The input gate formula is: in, The input gate weight matrix, For input gate bias terms; tanh is the hyperbolic tangent activation function. This is the candidate cell state weight matrix. This refers to the candidate cell state bias term; The cell state update formula is: ， in, This represents the element-wise product (Hadamard product). The output gate formula is , This is the output gate weight matrix. (for output gate bias term) ( (Hidden state at the current moment).

5. The clinical closed-loop nursing management method based on hospital chronic disease nursing big data according to claim 1, characterized in that, The anomaly score formula for the isolated forest anomaly detection model is as follows: ， Where s(x,n) is the anomaly score of sample x, n is the total number of samples, E(h(x)) is the average path length of sample x in all isolated trees in the isolated forest, and c(n) is the average path length of n samples (normalization factor).

6. The clinical closed-loop nursing management method based on hospital chronic disease nursing big data according to claim 4, characterized in that, The LSTM risk prediction model in S2 is trained using a federated learning algorithm, specifically the FedAvg algorithm, whose steps include: S21. The server initializes global model parameters; S22. Each client (hospital) downloads the global model parameters, trains the model using local chronic disease care data, and obtains the local model parameters; S23. Each client sends its local model parameters to the server; S24. The server averages the local model parameters of all clients to obtain new global model parameters; S25. Repeat steps S22-S24 until the model converges.

7. A clinical closed-loop nursing management device based on hospital chronic disease nursing big data, wherein the clinical closed-loop nursing management device based on hospital chronic disease nursing big data is used to implement the clinical closed-loop nursing management method based on hospital chronic disease nursing big data as described in any one of claims 1-6, characterized in that, The device includes: Data acquisition module: Used to collect patients' chronic disease care data and perform preprocessing; Risk assessment module: used to input pre-processed chronic disease care data into a pre-trained LSTM risk prediction model and output the patient's risk level of chronic disease exacerbation; Intervention plan generation module: used to generate personalized nursing intervention plans based on the risk level, combined with the patient's individual characteristics and historical intervention effect data; Execution tracking module: used to collect patients' health data in real time through wearable devices, input it into the isolated forest anomaly detection model, output anomaly scores and trigger alarms; The effect evaluation module is used to collect effect data on patients after receiving intervention and to evaluate the effectiveness of the intervention plan using the PSM model. The iterative optimization module is used to feed back the effect evaluation results to the risk assessment module and the intervention plan generation module, and update the model parameters and rules.

8. An electronic device, characterized in that, The electronic device includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 6.

Images