Stress damage assessment opportunity prediction method based on Braden scoring system and RNN model

Abstract

The invention relates to a stress injury assessment opportunity prediction method based on a Braden scoring system and an RNN model. The method comprises the following steps: collecting a Braden index and other related data of a patient; preprocessing the data; performing feature extraction on the collected data, and constructing a feature vector; an RNN model is trained through the feature vectors, and a stress damage risk prediction model is formed; inputting the real-time data of the patient into the trained stress injury risk prediction model, predicting the risk probability of the patient, and automatically determining the optimal time of the next evaluation according to a preset threshold value; and result output and feedback: feeding back the prediction result and the suggested evaluation opportunity to nursing personnel, and providing personalized nursing suggestions. According to the method, the risk change of the pressure injury of the patient can be automatically judged, the optimal evaluation and intervention opportunity is determined, the nursing quality is improved, and the occurrence rate of the pressure injury is reduced.

Description

Technical Field

[0001] The present invention relates to the field of medical care, and in particular to a method for predicting the timing of pressure injury assessment based on the Braden scoring system and an RNN model. Background Art

[0002] Pressure injuries are common complications in patients with long-term bed rest, limited mobility or sensory impairment, which seriously affect the quality of life of patients and increase medical costs. The Braden scoring system is a widely used clinical risk assessment scale for pressure injuries, which includes six dimensions: sensory perception, moisture level, activity, mobility, nutritional status and friction / shear force. However, the traditional Braden score mainly relies on the subjective assessment of nursing staff, which has the problems of low assessment frequency, poor timeliness and strong subjectivity.

[0003] For example, when monitoring high-risk patients who are bedridden for a long time, have limited mobility or sensory impairment, they usually need to undergo a Braden assessment every 48 hours. The average number of assessments per person is as high as 15-18 times within 30 days, and each patient needs to be assessed almost every 2 days, which is very time-consuming and labor-intensive. According to existing statistical data, the recognition rate of high-risk patients assessed by subjective methods is only about 50% of the actual number of high-risk patients, and the error is significantly large. Therefore, the incidence of pressure injuries also exceeds 10%, which is a relatively high frequency of occurrence. All these data and phenomena show that traditional assessment methods not only consume more manpower, but also have relatively obvious errors.

[0004] With the development of medical information technology and artificial intelligence technology, the use of Braden scoring indicators and patients' real-time data to build an automated and intelligent pressure injury risk assessment model has important research and application value. Summary of the invention

[0005] The present invention provides a method for predicting the timing of pressure injury assessment based on the Braden scoring system and the RNN model (recurrent neural network model). By comprehensively collecting various indicators in the Braden score and related physiological, behavioral, and environmental data, the recurrent neural network model automatically determines the risk changes of pressure injuries in patients, determines the optimal assessment and intervention timing, improves the quality of nursing, and reduces the incidence of pressure injuries.

[0006] The method for predicting the timing of pressure injury assessment based on the Braden scoring system and the RNN model of the present invention comprises the following steps: Data collection: According to the Braden scoring system, the six scoring indicators of the patient's Braden score, as well as physiological indicators, behavioral data, environmental data and historical medical records are collected. The behavioral data is obtained by using an electromyographic sensor to obtain the electromyographic signal generated by the patient's body movements, and then the electromyographic signal is converted into corresponding data representing the behavior through time domain conversion method or frequency domain conversion method; Data preprocessing: including data cleaning, coding, standardization and time alignment of collected data; Feature extraction and construction: Extract features from the collected Braden score indicators and other related data to construct feature vectors; Model training: training a recurrent neural network model (RNN model) through the feature vector to form a pressure injury risk prediction model; Risk prediction and assessment timing determination: Input the patient's real-time data into the trained pressure injury risk prediction model to predict the patient's risk probability and automatically determine the best time for the next assessment based on the preset threshold; Result output and feedback: The predicted results and recommended assessment timing are fed back to nursing staff to provide personalized nursing recommendations.

[0007] Furthermore, the physiological indicators in the data collection step include: body temperature, heart rate, blood pressure, blood oxygen saturation and blood sugar level.

[0008] Furthermore, the behavioral data in the data collection step includes: frequency of body position changes and amount of activity.

[0009] Furthermore, the environmental data in the data collection step includes: ward temperature and ward humidity.

[0010] Furthermore, in the feature extraction and construction step, the constructed feature vector includes the standardized values ​​of the six scoring indicators of the Braden score and other related data.

[0011] Furthermore, in the model training step, the update formula of the hidden state of the hidden layer in the recurrent neural network model is: ; Output pressure injury risk : ; in: W xh : The weight matrix input to the hidden layer; W hh : The weight matrix from hidden layer to hidden layer reflects the recursive relationship in time; W hy : The weight matrix from the hidden layer to the output layer; b h : The bias term of the hidden layer; b y : Bias vector of the output layer; tanh: hyperbolic tangent activation function; : Sigmoid activation function, mapping the output to between (0,1).

[0012] Furthermore, the loss function for training the recurrent neural network model uses the binary cross entropy loss function: ; in, N is the sample size, y i is the true label of whether pressure injury occurs. The risk probability predicted by the model.

[0013] Furthermore, in the risk prediction and determination of the timing of the evaluation, the method for determining the timing of the evaluation is: when the risk probability predicted by the pressure injury risk prediction model is greater than a preset threshold, it is recommended to conduct an immediate evaluation; when the risk probability is less than the threshold, but it is predicted that the risk probability will exceed the threshold within the future T time, then the evaluation is arranged before T time.

[0014] Furthermore, in the result output and feedback step, after the nursing staff evaluates the patient according to the predictions and suggestions of the pressure injury risk prediction model, the evaluation results and the nursing measures taken are fed back to the pressure injury risk prediction model for continuous learning and optimization of the pressure injury risk prediction model.

[0015] The beneficial effects of the present invention include: 1. Comprehensive data collection: Not only the Braden score indicators are collected, but also physiological, behavioral, environmental data and historical medical records are comprehensively considered to make risk assessment more accurate and comprehensive.

[0016] 2. Unique data preprocessing method: Specific missing value processing, outlier processing, data coding and standardization methods are used to ensure data quality and effectively improve the reliability of the model.

[0017] 3. Innovation in feature vector construction: Integrate multiple data into specific feature vectors to provide richer information for model input.

[0018] 4. Model selection and training: Selecting the RNN model and combining it with a specific algorithm formula and optimizer for training can better process time series data and predict risk changes in real time.

[0019] 5. Automatic assessment timing determination: By setting thresholds and predicting future risk probabilities, the assessment timing is automatically determined, which optimizes the nursing process and improves work efficiency.

[0020] 6. Personalized nursing advice and feedback mechanism: Provide customized nursing measures according to different patient conditions, and continuously learn and optimize the model through the feedback mechanism, effectively improving the prediction performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The present invention is a flowchart of a method for predicting the timing of pressure injury assessment based on the Braden scoring system and the RNN model.

[0022] Figure 2 Schematic diagram of the patient's electromyographic signal. DETAILED DESCRIPTION

[0023] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application. Example

[0024] like Figure 1 As shown, the method for predicting the timing of pressure injury assessment based on the Braden scoring system and the RNN model of the present invention comprises the following steps: Data collection: According to the Braden scoring system, the six scoring indicators of the patient's Braden score, as well as physiological indicators, behavioral data, environmental data and historical medical records are collected. The behavioral data is obtained by obtaining the electromyographic signals generated by the patient's body movements through electromyographic sensors, and then the electromyographic signals are converted into corresponding data representing the behavior through time domain conversion or frequency domain conversion.

[0025] Myoelectric sensors (sEMG) can be attached to the surface of the skin to detect the electrical activity generated by muscle contraction during any action of the patient's body. For example, Delsys, Vicon, OptiTrack and other myoelectric sensors / systems can be used, which can obtain the changes in muscle electrical signals in real time and provide raw data for subsequent conversion. The induction accuracy of myoelectric sensors is high and can accurately capture subtle changes in myoelectric signals.

[0026] The time domain conversion method can be adopted in two ways: (1) Integrated electromyographic value (iEMG): It is calculated by integrating the electromyographic signal over a period of time. The formula is: Calculate, where EMG ( t ) is a function of the change of the electromyographic signal over time t, where t1 and t2 are the start and end times of the electromyographic signal. The iEMG value is related to the intensity and duration of muscle contraction when the patient's body is performing an action, and can be used as an indicator to measure the amount of activity. For example, when a patient bends his arm while drinking water, the iEMG value of the biceps brachii will increase with the increase in the number of bends and the degree of force applied each time.

[0027] (2) Root mean square (RMS): By formula: Calculate. Among them, EMG i is a discrete sample of myoelectric signal, n is the number of samples of myoelectric signal, and RMS can reflect the energy of myoelectric signal and is related to the contraction force of muscle. When the intensity of muscle activity increases, the RMS value will increase. When analyzing the amount of exercise activity, the amount of activity can be evaluated by comparing the changes in RMS values ​​in different time periods.

[0028] The frequency domain conversion method can analyze patient behavior using power spectral density (PSD): By formula: , perform fast Fourier transform (FFT) on the electromyographic signal, convert it from the time domain to the frequency domain, and obtain the power spectrum density function. It is an electromyographic signal The Fourier transform of f It is frequency. PSD can show the energy distribution of electromyographic signals at different frequencies. When the patient performs behavioral activities to cause muscle contraction, the proportion of high-frequency components will change. By analyzing the changes in specific frequency bands in the PSD curve (such as the mid-frequency band, which is usually related to muscle force generation), the frequency characteristics of muscle activity can be understood. For example, when the muscle is fatigued, the PSD curve will move toward the low frequency direction, which can be used as a basis for evaluating the frequency of exercise and the degree of fatigue.

[0029] When using machine algorithm models to analyze motion behavior, two algorithms can be used: (1) Classification algorithm: The collected EMG signals are used as input, and different movement patterns and frequencies are identified by training classification models (such as support vector machines, neural networks, etc.). First, a large amount of labeled EMG signal data, such as arm swing EMG signals at different movement frequencies, is collected as a training set. The trained model can classify the newly collected EMG signals and determine the movement frequency category to which they belong.

[0030] (2) Regression algorithm: used to establish the relationship between EMG signal features and movement activity (such as joint angle change, number of steps, etc.). For example, a linear regression or nonlinear regression model is used, with EMG signal features such as RMS and iEMG as independent variables and actual measured movement activity as the dependent variable for training. After training, the model can determine the patient's behavioral activity based on the new EMG signal features.

[0031] Braden's scoring system includes the following six dimensions, each with a score range of 1-4: (1) Sensory Perception (SP): Assess the patient's level of discomfort in terms of pain and pressure.

[0032] 1 point: completely restricted; 2 points: severely limited; 3 points: slightly limited; 4 points: No restrictions.

[0033] (2) Moisture (M): Assess the degree of skin exposure to a moist environment.

[0034] 1 point: continuous moisture; 2 points: very moist; 3 points: occasionally moist; 4 points: Slightly moist.

[0035] (3) Activity (A): Assess the patient's activity level.

[0036] 1 point: bedridden; 2 points: confined to the chair; 3 points: walking occasionally; 4 points: Walk frequently.

[0037] (4) Mobility (Mob): Assess the patient's ability to change and control body position.

[0038] 1 point: Not at all; 2 points: very limited; 3 points: slightly limited; 4 points: Unrestricted.

[0039] (5) Nutritional status (N): Assess the patient's nutritional intake.

[0040] 1 point: very poor; 2 points: may be insufficient; 3 points: sufficient; 4 points: Good.

[0041] (6) Friction and Shear (FS): Assess the degree of friction and shear on the skin.

[0042] 1 point: There are problems; 2 points: potential problem; 3 points: No obvious problems.

[0043] Other relevant data include: 1. Physiological indicators (obtained through monitors or wearable devices): (1) Body temperature: The normal range is approximately 36.1℃-37.2℃.

[0044] (2) Heart Rate: The normal range is approximately 60-100 beats / minute.

[0045] (3) Blood Pressure: Systolic and diastolic pressure, the normal range is approximately 90 / 60 mmHg - 120 / 80 mmHg.

[0046] (4) Blood oxygen saturation (SpO2): The normal range is approximately 95%-100%.

[0047] (5) Blood Glucose Level: Normal fasting blood glucose is approximately 3.9-6.1 mmol / L.

[0048] 2. Behavioral data (obtained through electromyographic sensors or nursing records): (1) Position Change Frequency: The number of times a patient changes his or her position within a certain period of time.

[0049] (2) Activity Level: Activity intensity measured by an accelerometer.

[0050] For example, the electromyographic signal of the patient can be obtained by an electromyographic sensor, such as Figure 2 Then, through the above method, Figure 2The electromyographic signals are converted into curves representing the amplitude and amount of movement. For example, the myoelectric neurons (MOTORNERVES) include the electromyographic values ​​of the ankle-knee and knee joint of the right tibia (Right Tibialis) and the left tibia (Left Tibialis), as well as the electromyographic values ​​of the ankle-knee and knee joint of the right peroneus (Right Peroneus) and the left peroneus (Left Peroneus). Thus, the activity curves of the right tibialis anterior (Right Tibialis anterior) and the left tibialis anterior (Left Tibialis anterior) are obtained to represent the amount of movement of the corresponding parts of the patient.

[0051] 3. Environmental data (obtained through environmental sensors): (1) Room Temperature: Ambient temperature affects skin condition.

[0052] (2) Room Humidity: Environmental humidity affects the degree of skin moisture.

[0053] 4. Historical medical records (obtained from the electronic medical record system): (1) Medical History: such as diabetes, cardiovascular disease, nervous system disease, etc.

[0054] (2) Medication: Medications currently in use, such as hormones, sedatives, etc.

[0055] (3) Allergies: Allergic reactions to drugs or substances.

[0056] Example: Patient A, male, 65 years old, suffered from sequelae of stroke, was bedridden for a long time, and had a history of diabetes.

[0057] Braden's Rating: SP: 2 points (severely limited); M: 3 points (occasionally moist); A: 1 point (bedridden); Mob: 2 points (very limited); N: 2 points (may be insufficient); FS: 1 point (there are problems); Physiological indicators: Body temperature: 36.5℃; Heart rate: 80 beats / minute; Blood pressure: 130 / 85 mmHg; Blood oxygen saturation: 96%; Blood sugar level: 8.5 mmol / L (high); Behavioral data: Frequency of position change: once every 4 hours; Environmental data: Ward temperature: 24℃; Humidity in ward: 60%; Medical History: Past medical history: diabetes, hypertension; Medication: hypoglycemic drugs, antihypertensive drugs.

[0058] Data preprocessing: including data cleaning, coding, standardization and time alignment of the collected data.

[0059] Data cleaning: Missing value processing: If an indicator is missing, the mean or median will be used to fill it, or a reasonable estimate will be made based on clinical judgment.

[0060] Outlier processing: Identify and process outliers in physiological indicators, such as heart rate that is too high (>100 beats / minute) or too low (<60 beats / minute), and adjust or eliminate them based on actual conditions.

[0061] Data encoding: Braden score encoding: The score of each dimension is directly used as a numerical feature without additional encoding.

[0062] Categorical variable encoding: For categorical variables such as medication status, one-hot encoding or label encoding is used.

[0063] Data Standardization: Normalize continuous numerical features (such as body temperature, heart rate, blood pressure, blood sugar level, etc.) using the formula: ; Among them, x is the original value, is the mean, is the standard deviation.

[0064] Time Alignment: Align the collection time of different data sources to ensure that data at the same moment matches.

[0065] Feature extraction and construction: The collected Braden score indicators and other related data are used for feature extraction to construct feature vectors.

[0066] Constructing the feature vector of model inputX t , including Braden's score indicators and other relevant data: .

[0067] in: SP t : sensory perception score; M t : Humidity rating; A t : Activity score; Mob t : Mobility score; N t : Nutritional status score; FS t : Friction / shear score; T t :body temperature; HR t : heart rate; : systolic blood pressure; : Diastolic blood pressure; SpO2 t : Blood oxygen saturation; BG t : Blood sugar level; PCF t : Frequency of body position changes; AL t : Activity level; RT t : ward temperature; RH t : Humidity in the ward; MH t : Past medical history (after coding); MED t : Medication status (after coding); Example: The feature vector of patient A at time t is: .

[0068] Model training: a recurrent neural network model (RNN model) is trained by the feature vector to form a pressure injury risk prediction model.

[0069] (1) Model selection: Recurrent Neural Network (RNN): Suitable for processing time series data and can capture the dynamic changes of features over time.

[0070] (2) Model structure: Input layer: accepts feature vector X t .

[0071] Hidden layer: RNN unit, including hidden state h t , which can receive the hidden state from the previous time step. The hidden state is the internal state used by the RNN model to store and transfer information when processing the input sequence. It is updated at each time step to capture the information and context in the input sequence.

[0072] Output layer: Output risk probability y through activation function t .

[0073] (3) Algorithm formula: Hide status updates: ; Output pressure injury risk : ; in: W xh : The weight matrix input to the hidden layer; W hh : The weight matrix from hidden layer to hidden layer reflects the recursive relationship in time; W hy : The weight matrix from the hidden layer to the output layer; b h : The bias term of the hidden layer; b y : Bias vector of the output layer; tanh: hyperbolic tangent activation function; : Sigmoid activation function, mapping the output to between (0,1).

[0074] At the beginning of model training, each weight matrix is ​​usually randomly initialized. Common initialization methods include: Uniform distribution random initialization: randomly generated within a certain range.

[0075] Normal distribution random initialization: randomly generated according to mean and variance.

[0076] He initialization and Xavier initialization: Calculate the appropriate initial value range based on the number of network layers and activation function type.

[0077] Model training process: Forward Propagation: Input feature vector X t, calculate the hidden state h by the above formula t and the risk probability of output .

[0078] Calculate the loss function: Using the actual target values ​​(true labels) and the predicted output of the model, the loss is calculated.

[0079] Loss function (Binary Cross-Entropy Loss): ; : The true label (1 if pressure injury occurs, 0 if not).

[0080] : The risk probability predicted by the model, which represents the probability of a patient developing a pressure injury at time step t.

[0081] Backward Propagation Through Time (BPTT): Calculate the gradient (partial derivative) of the loss function with respect to each weight matrix.

[0082] Purpose: To determine the direction and magnitude of adjustment required for each parameter to reduce losses.

[0083] Parameter update: Update the weight matrix using an optimization algorithm such as the Adam optimizer.

[0084] Update formula (taking gradient descent as an example): ; : Model parameters, including the following parts: W xh : The weight matrix from the input layer to the hidden layer; W hh : The weight matrix from hidden layer to hidden layer (reflecting the recursive relationship in time); W hy : The weight matrix from the hidden layer to the output layer; b h : The bias term of the hidden layer; b y : The bias term of the output layer; : Learning rate, used to control the step size of model parameter update; : The gradient of the loss function with respect to the parameters, which indicates the rate of change of the loss function relative to the model parameters.

[0085] (4) Loss function: To calculate the loss function L, we need to first get the output of the model, that is, the risk probability The risk probability is obtained through the model The steps are: 1. Model output process: (1) Input data: For each sample i, the input feature is X i , which contains patient-related metrics (such as Braden score and other clinical data).

[0086] (2) Forward propagation calculation: Hide status updates:

[0087] Here, h t is the hidden state at the current time step, which combines the current input and the hidden state at the previous time step.

[0088] Output calculation: ; By using the Sigmoid activation function σ, the hidden state h t Map to (0,1)(0, 1)(0,1) to get the risk probability .

[0089] 2. Calculation of loss function: The output risk probability of the model is obtained After , it can be used to calculate the loss function L. For binary classification problems (such as predicting whether pressure injury occurs), the binary cross entropy loss function is usually used to measure the difference between the predicted value and the true label.

[0090] Define the binary cross entropy loss function: ; Where N is the number of samples, y i is the true label of whether pressure injury occurs (1 if pressure injury occurs, 0 if not), The risk probability predicted by the model.

[0091] The significance of loss function: When the model's predicted value The closer to the true value , the smaller the value of the loss function L; otherwise, the value of the loss function L will increase.

[0092] (5) Optimization algorithm: The Adam optimizer is used: an adaptive learning rate optimization algorithm, and the learning rate is set to 0.001.

[0093] (6) Training process: Dataset partitioning: Divide the data into training set, validation set, and test set, for example, 8:1:1.

[0094] Model training: Train the model on the training set and use the validation set to adjust the hyperparameters.

[0095] Model evaluation: Evaluate the performance of the model on the test set. Indicators include accuracy, precision, recall, F1-score (an indicator used in statistics to measure the accuracy of a binary classification model. It is the harmonic mean of precision and recall, with a maximum value of 1 and a minimum value of 0), AUC (the area under the ROC curve and the coordinate axis), etc.

[0096] Example: The historical data of 1000 patients were used for training, of which 800 were training sets, 100 were validation sets, and 100 were test sets.

[0097] Risk prediction and assessment timing determination: Input the patient's real-time data into the trained pressure injury risk prediction model to predict the patient's risk probability and automatically determine the best time for the next assessment based on the preset threshold.

[0098] (1) Risk prediction: Real-time prediction: The feature vector X collected in real time t Input the trained RNN model to get the risk probability .

[0099] (2) Determine the timing of the assessment: Set threshold θ: Set the risk probability threshold according to clinical needs, for example, θ = 0.7.

[0100] Timing judgment: a. If , an immediate pressure injury evaluation is recommended.

[0101] b. If , but if it is predicted that the risk probability will exceed the threshold within the next T time, an assessment will be arranged in advance.

[0102] Predicting future risks: Using the model to predict the risk probability at T moments in the future .

[0103] Example: Patient A's current risk probability , exceeding the threshold of 0.7, the system recommends immediate evaluation.

[0104] Result output and feedback: The predicted results and recommended assessment timing are fed back to nursing staff to provide personalized nursing recommendations.

[0105] Output: Risk Probability: Current risk probability .

[0106] Risk trend: changes in future risk probability.

[0107] Assessment recommendations: whether an immediate assessment is needed, or when the next assessment should be scheduled.

[0108] Nursing recommendations: Provide targeted nursing measures based on the high-risk factors identified by the model.

[0109] Notification method: Notify caregivers via care management system, mobile app or text message.

[0110] Feedback Mechanism: After the assessment, the nursing staff will feed back the results (such as whether pressure injury has occurred and the nursing measures taken) to the system for continuous learning and optimization of the model.

[0111] Example: The system prompts the caregiver: "Patient A is currently at high risk of pressure injury (75%), and immediate assessment is recommended. High-risk factors include severely limited sensory perception, low mobility, and high blood sugar levels. It is recommended to strengthen body position changes, skin care, and blood sugar management."

[0112] Example 1: Risk prediction for stroke patients: Patient Information: Patient B, male, 70 years old, suffered from sequelae of stroke, long-term bedridden, and sensory impairment.

[0113] Data collection: Braden's Rating: SP: 1 point (completely restricted); M: 2 points (very moist); A: 1 point (bedridden); Mob: 1 point (absolutely impossible); N: 2 points (may be insufficient); FS: 1 point (there are problems).

[0114] Physiological indicators: Body temperature: 37.0℃; Heart rate: 90 beats / minute; Blood pressure: 140 / 90 mmHg; Blood oxygen saturation: 94%; Blood glucose level: 7.8 mmol / L.

[0115] Behavioral data: Frequency of position changes: Every 6 hours (lower).

[0116] Environmental data: Ward temperature: 25℃; Ward humidity: 65%.

[0117] Medical History: Past medical history: hypertension, stroke; Medication: antihypertensive drugs, anticoagulants.

[0118] Model predictions and results: Calculate the eigenvector X t , input the model and get the risk probability .

[0119] because , the system recommends immediate evaluation.

[0120] Nursing interventions: The nurse conducted a comprehensive assessment of Patient B and found signs of redness and swelling on the skin.

[0121] Measures to be taken: Increase the frequency of body position changes (once every 2 hours).

[0122] Use a pressure-relieving mattress.

[0123] Strengthen skin care and keep dry skin away.

[0124] Adjust nutritional intake and provide a high-protein diet.

[0125] Effect feedback: After one week, the patient's risk probability dropped to .

[0126] The present invention fully utilizes various indicators of Braden score and physiological, behavioral and environmental data of patients through comprehensive data collection to improve the accuracy of risk assessment.

[0127] And through clear algorithm formulas and model structures, nursing staff can understand the decision-making basis of the model and enhance trust.

[0128] The time series processing capability of the RNN model is also used to predict risk changes in real time and take timely intervention measures. Customized nursing measures are provided according to the specific conditions of different patients to improve nursing effects. Finally, through the feedback mechanism, the model can continuously learn new data and continuously improve prediction performance.

[0129] Embodiment 2: On the basis of Example 1, the superiority of the present invention is further illustrated by comparing the data with the traditional prediction method.

[0130] Subjects: Patient population: 200 patients, all of whom are in high-risk groups, such as those who are bedridden for a long time, have limited mobility or sensory impairment.

[0131] Grouping: Experimental group (the present invention): 100 patients, using the pressure injury assessment timing prediction method based on the Braden scoring system and the RNN model of the present invention.

[0132] Control group (traditional method): 100 patients, evaluated using the traditional regular Braden score (eg, every 48 hours).

[0133] Experimental period: Duration: One month (30 days) Frequency of assessment: Experimental group: Real-time monitoring through the corresponding system, intelligent reminders for nursing staff to conduct assessments.

[0134] Control group: Evaluation was performed every 48 hours as usual.

[0135] Experimental results: Table 1: Number of evaluations and evaluation intervals index Experimental group (the present invention) Control group (traditional method) Total evaluation times 592 1512 Number of assessments per capita 6 15 Average time between evaluations 5 days 2 days Table 2: Identification rate of high-risk patients index Experimental group (the present invention) Control group (traditional method) Number of high-risk patients identified 82 59 High-risk patient identification rate 82% 59% Table 3: Number of nursing interventions index Experimental group (the present invention) Control group (traditional method) Total number of nursing interventions 421 306 Number of interventions per capita 4 3 Table 4: Incidence of pressure injuries index Experimental group (the present invention) Control group (traditional method) Number of patients with pressure injuries 5 13 The incidence of pressure injuries 5% 13% The results are analyzed through Tables 1 to 4: 1. Improved evaluation efficiency: The average number of evaluations per person in the experimental group was 6, which was less than the 15 in the control group.

[0136] The average interval between assessments was 5 days, which was longer than 2 days in the control group.

[0137] The number of assessments was reduced by 60%, but the effect on the prevention of pressure injuries was better, indicating that the present invention can optimize the timing of assessments and improve efficiency.

[0138] 2. Improved identification rate of high-risk patients: The experimental group identified 80% of high-risk patients, higher than the 60% in the control group.

[0139] Timely identification of high-risk patients can help take targeted nursing measures to reduce the risk of pressure injuries.

[0140] 3. Nursing intervention is more effective: The average number of interventions per person in the experimental group was 4 times, slightly higher than the 3 times in the control group.

[0141] Intervention measures are more targeted, focusing on patients who really need them, and improving the efficiency of nursing resources utilization.

[0142] 4. Reduced incidence of pressure injuries: The incidence of pressure injury in the experimental group was 5%, which was significantly lower than 15% in the control group.

[0143] The incidence of pressure injuries was reduced by about 66.7%, indicating that the system of the present invention is more effective in preventing pressure injuries.

[0144] It can be seen from the above data and analysis results that, compared with traditional methods, the present invention can predict risk changes in real time and take timely and effective intervention measures, which significantly improves the timeliness and effectiveness of nursing care.

[0145] The above-mentioned embodiments only express the specific implementation methods of the present application, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the protection scope of the present application. It should be pointed out that for ordinary technicians in this field, relevant deformations and improvements can be made without departing from the technical solution concept of the present application, and these all belong to the protection scope of the present application.

Claims

1. A method for predicting the timing of pressure injury assessment based on the Braden scoring system and RNN model. The method is characterized in that it comprises the steps of: Data collection: According to the Braden scoring system, the six scoring indicators of the patient's Braden score, as well as physiological indicators, behavioral data, environmental data and historical medical records are collected. The behavioral data is obtained by using an electromyographic sensor to obtain the electromyographic signal generated by the patient's body movements, and then the electromyographic signal is converted into corresponding data representing the behavior through time domain conversion method or frequency domain conversion method; Data preprocessing: including data cleaning, coding, standardization and time alignment of collected data; Feature extraction and construction: Extract features from the collected Braden score indicators and other related data to construct feature vectors; Model training: training a recurrent neural network model through the feature vector to form a pressure injury risk prediction model; Risk prediction and assessment timing determination: Input the patient's real-time data into the trained pressure injury risk prediction model to predict the patient's risk probability and automatically determine the best time for the next assessment based on the preset threshold; Result output and feedback: The predicted results and recommended assessment timing are fed back to nursing staff to provide personalized nursing recommendations.

2. The method for predicting the timing of pressure injury assessment according to claim 1, characterized in that: The physiological indicators in the data collection step include: body temperature, heart rate, blood pressure, blood oxygen saturation and blood sugar level.

3. The method for predicting the timing of pressure injury assessment according to claim 1, characterized in that: The behavioral data in the data collection step include: frequency of body position changes and activity amount.

4. The method for predicting the timing of pressure injury assessment according to claim 1, characterized in that: The environmental data in the data collection step includes: ward temperature and ward humidity.

5. The method for predicting the timing of pressure injury assessment according to claim 1, characterized in that: In the feature extraction and construction steps, the constructed feature vector contains the standardized values ​​of the six scoring indicators of Braden score and other related data.

6. The method for predicting the timing of pressure injury assessment according to claim 1, characterized in that: In the model training step, the update formula for the hidden state of the hidden layer in the recurrent neural network model is: ； Output pressure injury risk : ； in: W xh : The weight matrix input to the hidden layer; W hh : The weight matrix from hidden layer to hidden layer reflects the recursive relationship in time; W hy : The weight matrix from the hidden layer to the output layer; b h : The bias term of the hidden layer; b y : Bias vector of the output layer; tanh: hyperbolic tangent activation function; : Sigmoid activation function, mapping the output to between (0,1).

7. The method for predicting the timing of pressure injury assessment according to claim 1, characterized in that: The loss function when training the recurrent neural network model uses the binary cross entropy loss function: ； in, N is the sample size, y i is the true label of whether pressure injury occurs. The risk probability predicted by the model.

8. The method for predicting the timing of pressure injury assessment according to claim 1, characterized in that: In the risk prediction and assessment timing determination, the method for determining the assessment timing is: when the risk probability predicted by the pressure injury risk prediction model is greater than a preset threshold, it is recommended to conduct an immediate assessment; when the risk probability is less than the threshold, but it is predicted that the risk probability will exceed the threshold within the future T time, then arrange the assessment before T time.

9. The method for predicting the timing of pressure injury assessment according to claim 1, characterized in that: In the result output and feedback step, the nursing staff evaluates the patient according to the predictions and recommendations of the pressure injury risk prediction model, and then feeds back the evaluation results and nursing measures taken to the pressure injury risk prediction model for continuous learning and optimization of the pressure injury risk prediction model.

Images