Enteral nutritional complication risk grading early warning management method and related device

Abstract

The invention discloses an enteral nutritional complication risk grading early warning management method and a related device. The method comprises the following steps: generating a multi-modal data vector according to clinical data, environmental data and behavior data of a patient; inputting the multi-modal data vector and the individual feature information of the patient into a trained dynamic risk prediction model, and outputting occurrence probabilities of gastrointestinal complications, mechanical complications, infectious complications and metabolic complications; wherein the individual feature information comprises age, basic diseases and nutrition states; generating a personalized intervention scheme of the patient according to a risk grading result for reference of medical staff; and inputting the intervention effect of the medical personnel, the real-time data feedback of the patient and the enteral nutritional complication risk early warning system into the dynamic risk prediction model so as to adjust model parameters of the dynamic risk prediction model. Through the dynamic risk prediction model and the personalized intervention scheme, the prediction precision of the enteral nutritional complication risk is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of risk prediction and graded management of enteral nutrition complications, and in particular to a method and device for graded early warning management of enteral nutrition complications risk, as well as computing equipment. Background Art

[0002] Enteral nutrition refers to a nutritional support method that provides the gastrointestinal tract with metabolically necessary nutrients and other nutrients through oral or tube feeding. Compared to parenteral nutrition, it is more consistent with human physiology and plays an important role in protecting the gastrointestinal mucosa and maintaining intestinal ecological balance. It has become the preferred nutritional support method for patients with preserved gastrointestinal function and is widely used in clinical practice. However, during clinical implementation, it can also be prone to gastrointestinal, mechanical, infectious, and metabolic complications, leading to interruptions in enteral nutrition, inadequate feeding, prolonged hospitalization, and increased hospital costs. Currently, the incidence of aspiration during enteral nutrition can reach 30%, increasing the risk of aspiration pneumonia by 12-fold and contributing to a mortality rate of up to 70% for acute respiratory distress syndrome. The incidence of hyperglycemia during enteral nutrition in ICU patients can reach 46%, and in general ward patients, it can reach 32%. The incidence of feeding intolerance during enteral nutrition in ICU patients ranges from 41.27% to 73.6%, with abdominal distension occurring most frequently in 68.52%, constipation in 46.30%, and diarrhea in 42.59%. This not only directly impacts patient experience, prognosis, and clinical outcomes, but also increases nursing workload due to the management of related symptoms. Ensuring the smooth implementation of enteral nutrition and reducing or avoiding related complications are key clinical concerns. Currently, healthcare professionals typically conduct risk assessments through regular ward rounds, patient observation, vital sign recording, and laboratory test results. This relies heavily on experience and subjective judgment, which can be prone to oversights and errors. Furthermore, monitoring is limited in frequency, making real-time monitoring impossible and difficult to detect early risk signals. Therefore, it is necessary to provide an early warning management system that can monitor the patient status in near real time and adjust the risk classification in a timely manner. Through real-time monitoring and intelligent analysis, a reminder function is set for early warning items that reach the early warning level, and the severity of the level is distinguished by color, thereby improving nursing efficiency and enhancing the quality of enteral nutrition complication prevention nursing.

[0003] To solve the above problems, the present invention proposes a risk classification and early warning management method for enteral nutrition complications, which improves the prediction accuracy of enteral nutrition complication risks through a dynamic risk prediction model and personalized intervention plan. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method and device for early warning management of enteral nutrition complication risk classification, and a computing device.

[0005] According to one aspect of the present invention, a method for risk classification and early warning management of enteral nutrition complications is provided, comprising:

[0006] Generate a multimodal data vector based on the patient's clinical data, environmental data, and behavioral data; wherein the clinical data includes blood glucose level, gastrointestinal function indicators, nutrient solution infusion rate and temperature, bowel movement diary, and Braden scale; the environmental data includes ward temperature and humidity and nutrient solution storage conditions; and the behavioral data includes changes in patient position and activity frequency;

[0007] The patient's multimodal data vector and individual characteristic information are input into a trained dynamic risk prediction model to output the probability of gastrointestinal complications, mechanical complications, infectious complications, and metabolic complications; wherein the dynamic risk prediction model includes an LSTM network layer, a Transformer network layer, an MLP network layer, a Stacking network layer, a clustering analysis layer, and a LIME network interpretation layer; the individual characteristic information includes age, underlying diseases, and nutritional status; the clustering analysis layer is used to perform cluster analysis based on the probability of occurrence of each complication output by the Stacking network layer, and output risk grading results including four levels of low risk, medium risk, high risk, and critical risk;

[0008] A personalized intervention plan for the patient is generated based on the risk grading results for reference by medical staff; the intervention effect of medical staff, the real-time data feedback of the patient and the enteral nutrition complication risk warning system are input into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model; wherein, the enteral nutrition complication risk warning system includes a proposed item pool, a first round of expert inquiries and a second round of evaluation results.

[0009] In an optional manner, the LSTM network layer includes a blood glucose trend analysis unit, a gastrointestinal function monitoring unit, a nutrient solution infusion optimization unit, an environmental factor influence unit, a body position and activity analysis unit, a metabolic state prediction unit, an infection risk assessment unit, a time series feature extraction unit, a dynamic risk integration unit, and a long-term dependency capture unit;

[0010] Wherein, the blood glucose trend analysis unit is used to analyze the trend of the patient's blood glucose level over time and predict its correlation with metabolic complications;

[0011] The gastrointestinal function monitoring unit is used to analyze the changing patterns of gastric emptying time and intestinal motility frequency and predict the risk of gastrointestinal complications;

[0012] The nutrient solution infusion optimization unit is used to monitor changes in nutrient solution infusion rate and temperature, and to predict the risk of diarrhea and abdominal distension in gastrointestinal complications;

[0013] The environmental factor impact unit is used to analyze the impact of ward temperature and humidity, and nutrient solution storage conditions on the patient's intestinal flora and infectious complications;

[0014] The body position and activity analysis unit is used to analyze the effects of changes in the patient's body position and activity frequency on gastrointestinal function, and to predict the risk of gastrointestinal complications caused by improper body position or insufficient activity;

[0015] The metabolic state prediction unit is used to integrate blood sugar level and nutrient solution infusion rate data and predict the risk of metabolic complications;

[0016] The infection risk assessment unit analyzes the risk of infectious complications based on the environmental data and behavioral data;

[0017] The time series feature extraction unit is used to integrate the time series features of clinical data, environmental data and behavioral data;

[0018] The dynamic risk integration unit is used to generate comprehensive risk prediction results;

[0019] The long-term dependency capture unit is used to capture the cumulative impact of a continuous increase in blood sugar levels on metabolic complications, or the potential threat of long-term abnormal temperature and humidity in the ward to infectious complications.

[0020] In an optional manner, the first round of expert inquiry and the second round of evaluation results further include:

[0021] Through literature review, clinical experience summary and patient interviews, we collected, screened and classified physiological factors, enteral nutrition factors, environmental factors and patient behavior factors related to the risk of enteral nutrition complications to form an initial pool of operation items;

[0022] Invite a preset number of experts in the field of enteral nutrition, provide the initial pool of items to the experts, have the experts perform a first round of scoring on the initial pool of items, and screen the pool of items according to preset scoring criteria;

[0023] The screened item pool is sent again to the same group of experts for a second round of evaluation. If the evaluation results of the second round are within the preset range of the evaluation results of the first round, the screened item pool reaches a stable state. The screened item pool is used as a component of the enteral nutrition complication risk warning system to assess the patient's risk factors.

[0024] In an optional manner, the Stacking network layer includes a basic model layer and a meta-learner layer; the LIME network interpretation layer includes a local perturbation module, a model prediction module, an interpreter module, a visualization module and a rule extraction module.

[0025] In an optional manner, generating a personalized intervention plan for the patient based on the risk grading result further includes:

[0026] Differentiated enteral nutrition support plans are developed for patients at different risk levels based on expert consensus, clinical guidelines, and individual patient conditions. For low-risk patients, formulas containing short peptides or medium-chain triglycerides are selected to promote absorption and reduce gastrointestinal burden. For medium-risk patients, digestion and absorption are assessed through abdominal auscultation and measurement of gastric residual volume, and the infusion rate and concentration of the nutrient solution are adjusted based on the monitoring results. For high-risk patients, nasojejunal or gastrojejunal tubes are used for infusion, bypassing the stomach to reduce the risk of aspiration.

[0027] For patients at different risk of complications, a small dose and slowly increasing infusion rate are used to avoid overburdening the gastrointestinal tract. For patients at risk of mechanical complications, the catheter position is regularly checked to ensure catheter patency. For patients at risk of metabolic complications, the nutrient solution formula and infusion rate are adjusted, and insulin medication is used to control blood sugar, with fine-tuning performed based on the rules provided by the LIME network interpretation layer.

[0028] The intervention plan is dynamically adjusted according to the patient's actual situation and intervention effect for reference by medical staff; if the patient has diarrhea, the concentration of the nutrient solution is reduced, the infusion rate is adjusted, or the nutrient solution formula is changed; if the patient's blood sugar continues to rise, the insulin dosage is adjusted or the nutrient solution type is changed.

[0029] In an optional manner, the training process of the dynamic risk prediction model further includes:

[0030] Divide the historical multimodal data vector into training set, validation set and test set;

[0031] The LSTM network layer is trained using the training set to learn long-term dependencies in time series data; the output of the LSTM network layer is used as the input of the Transformer network layer to learn the global relationship between the data; the output of the Transformer network layer and individual feature information are input into the MLP network layer for feature fusion and nonlinear transformation; the output of the MLP network layer is input into the Stacking network layer to perform ensemble learning using multiple basic models; and the number of hidden units, learning rate, and dropout rate of the LSTM network layer and the Transformer network layer are adjusted using the validation set;

[0032] After the dynamic risk prediction model is trained, the LIME network interpretation layer is used to interpret the prediction results. By making local perturbations near the prediction point and observing the changes in the prediction results, the importance of each feature to the prediction results is inferred.

[0033] In an optional manner, inputting the intervention effect of medical staff, the real-time data feedback of patients, and the enteral nutrition complication risk warning system into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model further includes:

[0034] Real-time collection of patient data during enteral nutrition, including daily blood sugar, blood gas analysis, electrolytes, renal function, liver function, and inflammatory indicators;

[0035] Automatically collect information on the type, rate, dosage, and route of nutrient solution by connecting to an infusion pump, blood glucose meter, and monitor; obtain information from medical staff on adjusting nutrient solution formulas, infusion rates, medications, and body position management, as well as the corresponding reasons for intervention, intervention measures, and expected effects; and record the patient's subjective experience and the time, frequency, and severity of symptoms.

[0036] If the patient's gastric residual volume decreases after adjusting the nutrient solution infusion rate, the intervention is effective; if the patient's diarrhea symptoms are alleviated after taking probiotics, the intervention is effective; based on the assessment results of the enteral nutrition complication risk warning system, determine whether the patient's risk level has changed; if the patient's risk level decreases, the intervention is effective; otherwise, the intervention needs to be adjusted;

[0037] The expected effects of medical staff, the index data of patients and the enteral nutrition complication risk warning system are input into the dynamic risk prediction model, and the model parameters are automatically adjusted based on the new data.

[0038] In an optional manner, the clustering formula of the cluster analysis layer is:

[0039] CorePoint=x i |(Neighborhood(x i ,ε)≥MinPts)∧(Density(x i ,ε,k)≥ρ threshold )∧(Distance(x i ,Centroid)≤δ×Distance Max )

[0040] Among them, CorePoint is the core point; x i Neighborhood(x i ,ε) is the value of the data point x iThe set of all data points whose distance is less than or equal to ε; ε is the neighborhood radius threshold; MinPts is the minimum number of neighbor points; Density(x i ,ε,k) is the data point x i The average density of k nearest neighbors within a radius of ε; k is the nearest neighbor parameter; ρ threshold is the density threshold; Distance(x i ,Centroid) is the data point x i The distance to the cluster center Centroid; δ is the distance coefficient; Distance Max is the data point x i The maximum distance to the cluster center to which it belongs.

[0041] According to another aspect of the present invention, a device for grading and warning the risk of enteral nutrition complications is provided, comprising:

[0042] a multimodal data vector generation module, configured to generate a multimodal data vector based on the patient's clinical data, environmental data, and behavioral data; wherein the clinical data includes blood glucose level, gastrointestinal function indicators, nutrient solution infusion rate and temperature, bowel movement diary, and Braden scale; the environmental data includes ward temperature and humidity and nutrient solution storage conditions; and the behavioral data includes changes in patient position and activity frequency;

[0043] A dynamic risk prediction module is used to input the patient's multimodal data vector and individual characteristic information into a trained dynamic risk prediction model, and output the probability of gastrointestinal complications, mechanical complications, infectious complications, and metabolic complications; wherein the dynamic risk prediction model includes an LSTM network layer, a Transformer network layer, an MLP network layer, a Stacking network layer, a clustering analysis layer, and a LIME network interpretation layer; the individual characteristic information includes age, underlying diseases, and nutritional status; the clustering analysis layer is used to perform cluster analysis based on the probability of occurrence of each complication output by the Stacking network layer, and output risk grading results including four levels of low risk, medium risk, high risk, and critical risk;

[0044] A personalized intervention plan generation module is used to generate a personalized intervention plan for the patient based on the risk grading results for reference by medical staff; the intervention effect of medical staff, the real-time data feedback of patients and the enteral nutrition complication risk warning system are input into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model; wherein, the enteral nutrition complication risk warning system includes a proposed item pool, the first round of expert inquiries and the second round of evaluation results.

[0045] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0046] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned enteral nutrition complication risk classification warning management method.

[0047] According to the solution provided by the present invention, a multimodal data vector is generated based on the patient's clinical data, environmental data and behavioral data; wherein, the clinical data include blood glucose level, gastrointestinal function index, nutrient solution infusion speed and temperature, bowel movement diary and Braden scale, the environmental data include ward temperature and humidity and nutrient solution storage conditions, and the behavioral data include patient posture changes and activity frequency; the patient's multimodal data vector and individual feature information are input into a trained dynamic risk prediction model, and the probability of occurrence of gastrointestinal complications, mechanical complications, infectious complications and metabolic complications is output; wherein, the dynamic risk prediction model includes LSTM network layer, Transformer network layer, MLP network layer, Stacking network layer, Layer, cluster analysis layer and LIME network interpretation layer; the individual characteristic information includes age, underlying diseases and nutritional status; the cluster analysis layer is used to perform cluster analysis based on the probability of occurrence of each complication output by the Stacking network layer, and output risk grading results including four levels of low risk, medium risk, high risk and critical risk; based on the risk grading results, a personalized intervention plan for the patient is generated for reference by medical staff; the intervention effect of medical staff, real-time data feedback of patients and enteral nutrition complication risk warning system are input into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model; wherein, the enteral nutrition complication risk warning system includes a proposed entry pool, a first round of expert consultation and a second round of evaluation results. The present invention generates multimodal data vectors based on clinical data, environmental data and behavioral data, which can more comprehensively reflect the patient's health status and external environmental factors. The cluster analysis layer is used to divide the patient's risk level into four levels of low risk, medium risk, high risk and critical risk, which facilitates medical staff to quickly understand the patient's risk status. Based on risk stratification results, expert consensus, and clinical guidelines, differentiated enteral nutrition support plans and interventions are developed to provide patients with more personalized medical services. The LIME network interpretation layer helps medical staff better understand risk factors and implement more refined interventions. Factors associated with the risk of enteral nutrition complications are collected through literature review, clinical experience summary, and patient interviews. Expert consultation and evaluation ensure the reliability of the early warning system.

[0048] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0050] Figure 1 A schematic diagram showing a process of a method for grading early warning management of enteral nutrition complication risks according to an embodiment of the present invention is shown;

[0051] Figure 2 A schematic diagram showing the workflow of the LIME network interpretation layer according to an embodiment of the present invention is shown;

[0052] Figure 3 A schematic diagram showing the process of training and parameter adjustment of a dynamic risk prediction model according to an embodiment of the present invention is shown;

[0053] Figure 4 A schematic diagram of an enteral nutrition approach according to an embodiment of the present invention is shown;

[0054] Figure 5 A schematic diagram showing a technical route for early warning of enteral nutrition complication risks according to an embodiment of the present invention is shown;

[0055] Figure 6 A schematic diagram showing a framework of a device for grading and early warning management of enteral nutrition complications risk according to an embodiment of the present invention is shown;

[0056] Figure 7 A schematic structural diagram of a computing device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0057] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0058] Figure 1 FIG. 1 is a flow chart showing a method for managing the risk classification of enteral nutrition complications according to an embodiment of the present invention. Figure 1 As shown, the following steps are included:

[0059] Step S101, generating a multimodal data vector based on the patient's clinical data, environmental data, and behavioral data; wherein the clinical data includes blood glucose level, gastrointestinal function indicators, nutrient solution infusion rate and temperature, bowel movement diary, and Braden scale; the environmental data includes ward temperature and humidity and nutrient solution storage conditions; and the behavioral data includes patient position changes and activity frequency.

[0060] In this embodiment, the patient's blood sugar level, gastrointestinal function indicators, nutrient solution infusion speed and temperature data, defecation diary (including stool shape, defecation frequency, etc., used to monitor the risk of diarrhea, abdominal distension and constipation) and Braden scale (skin risk assessment, used to assess the risk of nasogastric feeding patients suffering from nasogastric pressure injury) are collected in real time through blood glucose meters, gastrointestinal function monitoring equipment, infusion pumps and other equipment. The ward temperature and humidity and nutrient solution storage condition data are collected in real time through temperature and humidity sensors and nutrient solution storage monitoring equipment. The patient's body position changes and activity frequency data are collected in real time through smart mattresses. The above data are fused to obtain the following multimodal data vector: [130, 3, 75, 24, 25, 58, 4, 0, 2, 1, 18]. Among them, 130 represents blood glucose level (mg / dL), 3 represents gastrointestinal motility frequency (times / minute), 75 represents nutritional solution infusion rate (mL / h), 24 represents nutritional solution temperature (°C), 25 represents ward temperature (°C), 58 represents ward humidity (%RH), 4 represents nutritional solution storage temperature (°C), 0 represents body position (0 = bed rest, 1 = sitting, 2 = standing), 2 represents activity frequency (times / hour), 1 represents abdominal distension (0 = no, 1 = yes), which is obtained through a bowel diary and / or abdominal physical examination, and 18 represents the total score of the Braden scale, which is obtained through regular assessments.

[0061] Step S102: Input the patient's multimodal data vector and individual characteristic information into a trained dynamic risk prediction model to output the probability of occurrence of gastrointestinal complications, mechanical complications, infectious complications and metabolic complications; wherein the dynamic risk prediction model includes an LSTM network layer, a Transformer network layer, an MLP network layer, a Stacking network layer, a clustering analysis layer and a LIME network interpretation layer; the individual characteristic information includes age, underlying diseases and nutritional status; the clustering analysis layer is used to perform cluster analysis based on the probability of occurrence of each complication output by the Stacking network layer, and output risk grading results including four levels of low risk, medium risk, high risk and critical risk.

[0062] In this embodiment, the patient's multimodal data and individual characteristic information are used to accurately predict the probability of occurrence of complications after gastrointestinal surgery (gastrointestinal, mechanical, infectious, metabolic) and perform risk grading through a dynamic risk prediction model including LSTM, Transformer, MLP, Stacking, cluster analysis and LIME network interpretation layers, providing a basis for clinical decision-making. Specifically, through vital signs, laboratory tests, imaging data, and individual characteristic information such as age, underlying diseases and nutritional status, the patient's risk information can be captured more comprehensively. The LSTM network layer can process time series data and capture dynamic changes in the patient's disease course. The Transformer network layer learns the dependencies between different time points, thereby achieving dynamic risk prediction and providing a more timely basis for clinical intervention. The LSTM and Transformer network layers extract key features from time series data, the MLP network layer processes static individual characteristic information, and the Stacking network layer fuses the prediction results of different models to further improve the prediction accuracy. The cluster analysis layer converts continuous risk probabilities into easily understandable risk levels (low, medium, high, and critical), enabling a more intuitive understanding of a patient's risk level and the development of appropriate treatment plans. The LIME network interpretation layer explains the reasons behind the model's predictions, helping doctors understand which features play a key role in risk prediction, further promoting clinical application.

[0063] For example, consider a 65-year-old gastric cancer patient with hypertension, diabetes, a BMI of 22, and preoperative laboratory tests showing a low albumin level. The model inputs the patient's vital signs (blood pressure 150 / 90 mmHg, heart rate 90 beats / min), laboratory test results (albumin 30 g / L), imaging report, age 65, history of hypertension, diabetes, and BMI 22. After processing through the LSTM, Transformer, MLP, and Stacking layers, the model predicts a 60% probability of gastrointestinal complications, a 30% probability of mechanical complications, a 40% probability of infectious complications, and a 20% probability of metabolic complications. Based on these complication probabilities, the cluster analysis layer assigns the patient a "high risk" risk. The LIME interpretation layer indicates that hypertension, low albumin levels, and age are the primary factors contributing to the patient's "high risk" rating. Based on the model's predictions and the LIME interpretation, the doctor implements measures such as strengthening blood pressure control, improving nutritional status, and preventing infections to reduce the patient's risk of complications.

[0064] In this embodiment, the LSTM network layer includes a blood glucose trend analysis unit, a gastrointestinal function monitoring unit, a nutrient solution infusion optimization unit, an environmental factor influence unit, a body position and activity analysis unit, a metabolic state prediction unit, an infection risk assessment unit, a time series feature extraction unit, a dynamic risk integration unit, and a long-term dependency capture unit;

[0065] Wherein, the blood glucose trend analysis unit is used to analyze the trend of the patient's blood glucose level over time and predict its correlation with metabolic complications;

[0066] The gastrointestinal function monitoring unit is used to analyze the changing patterns of gastric emptying time and intestinal motility frequency and predict the risk of gastrointestinal complications;

[0067] The nutrient solution infusion optimization unit is used to monitor changes in nutrient solution infusion rate and temperature, and predict the risk of diarrhea and abdominal distension in gastrointestinal complications;

[0068] The environmental factor impact unit is used to analyze the impact of ward temperature and humidity, and nutrient solution storage conditions on the patient's intestinal flora and infectious complications;

[0069] The body position and activity analysis unit is used to analyze the effects of changes in the patient's body position and activity frequency on gastrointestinal function, and to predict the risk of gastrointestinal complications caused by improper body position or insufficient activity;

[0070] The metabolic state prediction unit is used to integrate blood sugar level and nutrient solution infusion rate data and predict the risk of metabolic complications;

[0071] The infection risk assessment unit analyzes the risk of infectious complications based on the environmental data and behavioral data;

[0072] The time series feature extraction unit is used to integrate the time series features of clinical data, environmental data and behavioral data;

[0073] The dynamic risk integration unit is used to generate comprehensive risk prediction results;

[0074] The long-term dependency capture unit is used to capture the cumulative impact of a continuous increase in blood sugar levels on metabolic complications, or the potential threat of long-term abnormal temperature and humidity in the ward to infectious complications.

[0075] In an optional manner, the Stacking network layer includes a basic model layer and a meta-learner layer; the LIME network interpretation layer includes a local perturbation module, a model prediction module, an interpreter module, a visualization module and a rule extraction module.

[0076] In this embodiment, Figure 2As shown in the figure, LIME provides explanations for specific samples by perturbing local areas. It not only provides visualization results of feature importance, but also extracts rules to help doctors understand the decision logic of the model.

[0077] In an optional manner, the training process of the dynamic risk prediction model further includes:

[0078] Divide the historical multimodal data vector into training set, validation set and test set;

[0079] The LSTM network layer is trained using the training set to learn long-term dependencies in time series data; the output of the LSTM network layer is used as the input of the Transformer network layer to learn the global relationship between the data; the output of the Transformer network layer and individual feature information are input into the MLP network layer for feature fusion and nonlinear transformation; the output of the MLP network layer is input into the Stacking network layer to perform ensemble learning using multiple basic models; and the number of hidden units, learning rate, and dropout rate of the LSTM network layer and the Transformer network layer are adjusted using the validation set;

[0080] After the dynamic risk prediction model is trained, the LIME network interpretation layer is used to interpret the prediction results. By making local perturbations near the prediction point and observing the changes in the prediction results, the importance of each feature to the prediction results is inferred.

[0081] In this embodiment, Figure 3As shown, the historical multimodal data vectors are divided into training, validation, and test sets in a ratio of 70%:15%:15%. The clinical, environmental, and behavioral data from the training set are fed into an LSTM network layer to learn long-term dependencies in the time series data (e.g., trends in blood glucose levels over time). This generates a time series feature vector, which serves as the input to the Transformer network layer. The LSTM network output is fed into the Transformer network layer to learn global dependencies between the data, generating a global feature vector as the input to the MLP network layer. The MLP network layer feeds the Transformer network output and individual feature information (e.g., age and underlying medical conditions) into the MLP network layer, where it performs feature fusion and nonlinear transformation to generate a high-dimensional feature vector. This is then fed into the Stacking network layer, where ensemble learning, such as using random forests or XGBoost, is used to generate the probabilities of gastrointestinal complications, mechanical complications, infectious complications, and metabolic complications. For example, to predict the risk of metabolic complications, patient A's persistently elevated blood glucose level (>180 mg / dL) and excessively rapid nutrient infusion (>100 mL / h) are input. The LSTM network layer analyzes the trend of blood sugar levels over time and generates a time series feature vector. The Transformer network layer analyzes the global dependency between blood sugar levels and the infusion rate of the nutrient solution and generates a global feature vector. The MLP network layer fuses the global feature vector with individual feature information (such as age and underlying diseases) to generate a high-dimensional feature vector. The Stacking network layer integrates the prediction results of multiple basic models and generates a 90% probability of metabolic complications. The LIME network interpretation layer extracts the rule "When the blood sugar level is >180mg / dL, the risk of metabolic complications increases significantly." It is recommended to adjust the nutrient solution formula (reduce the sugar content), reduce the infusion rate (<80mL / h), and use insulin to control blood sugar.

[0082] In an optional manner, the clustering formula of the cluster analysis layer is:

[0083] CorePoint=x i |(Neighborhood(x i ,ε)≥MinPts)∧(Density(x i ,ε,k)≥ρ threshold )∧(Distance(x i ,Centroid)≤δ×Distnace Max )

[0084] Among them, CorePoint is the core point; x i Neighborhood(xi ,ε) is the value of the data point x i The set of all data points whose distance is less than or equal to ε; ε is the neighborhood radius threshold; MinPts is the minimum number of neighbor points; Density(x i ,ε,k) is the data point x i The average density of k nearest neighbors within a radius of ε; k is the nearest neighbor parameter; ρ threshold is the density threshold; Distance(x i ,Centroid) is the data point x i The distance to the cluster center Centroid; δ is the distance coefficient; Distance Max is the data point x i The maximum distance to the cluster center to which it belongs.

[0085] In this embodiment, it is assumed that there are 1000 patients' complication probability vectors with a vector dimension of 5 (5 different complications). Select ε = 0.2, MinPts = 6, k = 5, ρ threshold =2.5, δ=0.75. Normalize the probability vector of complications for each patient. For the probability vector of patient A, calculate Neighborhood(x A ,0.2), find the distance x A For other patients less than or equal to 0.2, assuming there are 8, then Neighborhood(x A ,0.2)=8Neighborhood(x A ,0.2)≥MinPts(8≥6), which satisfies the condition. Calculate Density(x A ,0.2,5), find x A Calculate the 5 nearest neighbor patients of these patients to x A The reciprocal of the average distance, assuming the calculated result is 3.0. Density(x A ,0.2,5)≥Density(x A ,0.2,5)(3.0≥2.5), meets the condition. Assume that xA belongs to cluster C, calculate Distance(x A ,Centroid C ), assuming the distance is 0.1, calculate Distance Max Find all patients in cluster C to the cluster center Centroid C The maximum value of the distance is assumed to be 0.15. A ,Centroid C )≤δ×Distance Max(0.1≤0.75×0.15=0.1125), which meets the conditions. Since patient A meets all the conditions, it is determined to be a core point. Starting from patient A, expand to other patients in its neighborhood. If the patients in the neighborhood are also core points, continue to expand until it cannot be expanded. Divide all reachable core points and their neighborhood points into a cluster, and repeat the above steps until all core points have been visited. Through the above clustering process, patients are divided into different complication risk groups. For example, one cluster may represent patients with a high risk of cardiac complications, while another cluster may represent patients with a high risk of pulmonary complications. The above clustering results help doctors better understand the patient's risk situation and develop more targeted treatment plans.

[0086] Step S103, generating a personalized intervention plan for the patient based on the risk grading results for reference by medical staff; inputting the intervention effect of medical staff, the real-time data feedback of the patient and the enteral nutrition complication risk warning system into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model; wherein, the enteral nutrition complication risk warning system includes a proposed item pool, a first round of expert inquiries and a second round of evaluation results.

[0087] In this embodiment, Figure 5 As shown in the figure, by continuously inputting the intervention effects of medical staff, real-time data feedback from patients and the enteral nutrition complication risk warning system, the model parameters are dynamically adjusted so that the model can continue to learn and optimize. Among them, the proposed item pool is established based on historical data and expert experience, containing possible complications and corresponding warning conditions (for example, complications: intestinal perforation, warning conditions: the patient's abdominal pain lasts for more than 24 hours, accompanied by fever and diarrhea). The first round of expert consultation is to submit the proposed item pool to experts for preliminary evaluation to screen out high-risk complications and warning conditions. The results of the second round of evaluation are the warning conditions that are further evaluated and optimized based on the results of the first round of expert consultation.

[0088] In an optional manner, the first round of expert inquiry and the second round of evaluation results further include:

[0089] Through literature review, clinical experience summary and patient interviews, we collected, screened and classified physiological factors, enteral nutrition factors, environmental factors and patient behavior factors related to the risk of enteral nutrition complications to form an initial pool of operation items;

[0090] Invite a preset number of experts in the field of enteral nutrition, provide the initial pool of items to the experts, have the experts perform a first round of scoring on the initial pool of items, and screen the pool of items according to preset scoring criteria;

[0091] The screened item pool is sent again to the same group of experts for a second round of evaluation. If the evaluation results of the second round are within the preset range of the evaluation results of the first round, the screened item pool reaches a stable state. The screened item pool is used as a component of the enteral nutrition complication risk warning system to assess the patient's risk factors.

[0092] In this embodiment, the Delphi expert inquiry method was adopted to utilize the collective wisdom of experts and reduce subjective bias. Through two rounds of evaluation and setting consistency standards, the stability and consistency of item screening were ensured. For example, after literature review, clinical experience summary and patient interviews, the following factors that may be related to the risk of enteral nutrition complications were preliminarily collected, among which physiological factors were: age > 65 years old, diabetes, renal insufficiency, and immunodeficiency; enteral nutrition factors were: feeding speed too fast, hypertonic formula, and infusion tube contamination; environmental factors were: insufficient nursing staff, lack of standardized operating procedures, and poor ventilation in wards; patient behavior factors were: poor compliance, failure to report discomfort in a timely manner, and self-adjustment of feeding speed.

[0093] During the first round of expert consultation, the above items were sent to 15 enteral nutrition experts, who rated the importance and actionability of each item (1-5, with 1 being the least important / least actionable and 5 being the most important / most actionable). Assuming that, after statistical analysis, the experts rated the importance of "feeding too fast" as 4.5 points on average with a coefficient of variation of 0.1, meeting the pre-set screening criteria (average score > 4 points, coefficient of variation < 0.2), the item was retained. On the other hand, the importance of "poor ward ventilation" was 3.0 points on average with a coefficient of variation of 0.3, failing to meet the screening criteria, and the item was removed.

[0094] During the second round of expert consultation, the selected item pool and the statistical results from the first round were sent to the experts for a second round of scoring. The two rounds of scoring were compared. If the average score for each item changed by less than 0.5 points, the item pool was considered to have reached a stable state. The final selected items were incorporated into the enteral nutrition complication risk warning system to assess patient risk.

[0095] In an optional manner, generating a personalized intervention plan for the patient based on the risk grading result further includes:

[0096] Differentiated enteral nutrition support plans are developed for patients at different risk levels based on expert consensus, clinical guidelines, and individual patient conditions. For low-risk patients, formulas containing short peptides or medium-chain triglycerides are selected to promote absorption and reduce gastrointestinal burden. For medium-risk patients, digestion and absorption are assessed through abdominal auscultation and measurement of gastric residual volume, and the infusion rate and concentration of the nutrient solution are adjusted based on the monitoring results. For high-risk patients, nasojejunal or gastrojejunal tubes are used for infusion, bypassing the stomach to reduce the risk of aspiration.

[0097] For patients at different risk of complications, a small dose and slowly increasing infusion rate are used to avoid overburdening the gastrointestinal tract. For patients at risk of mechanical complications, the catheter position is regularly checked to ensure catheter patency. For patients at risk of metabolic complications, the nutrient solution formula and infusion rate are adjusted, and insulin medication is used to control blood sugar, with fine-tuning performed based on the rules provided by the LIME network interpretation layer.

[0098] The intervention plan is dynamically adjusted according to the patient's actual situation and intervention effect for reference by medical staff; if the patient has diarrhea, the concentration of the nutrient solution is reduced, the infusion rate is adjusted, or the nutrient solution formula is changed; if the patient's blood sugar continues to rise, the insulin dosage is adjusted or the nutrient solution type is changed.

[0099] In this embodiment, differentiated intervention plans are provided for patients with different risk levels and complication risks, and the reliability of the plans is guaranteed based on expert consensus and clinical guidelines. For example, for an elderly diabetic patient, after risk assessment, he was rated as medium risk and at risk of metabolic complications. The initial intervention plan selected an enteral nutrition formula containing short peptides, controlled the infusion rate to avoid being too fast, and used insulin to control blood sugar according to blood sugar monitoring results. LIME network interpretation layer analysis showed that the patient's blood sugar control was poor, and the most critical influencing factors were "carbohydrate intake" and "basal insulin dose". Under the advice of LIME, a low-sugar enteral nutrition formula was used and the basal insulin dose was increased. Closely monitor blood sugar changes and further adjust the insulin dosage based on blood sugar results. If the patient has diarrhea, reduce the concentration of the nutrient solution or switch to a lactose-free formula. If the patient's blood sugar is still poorly controlled, further adjust the insulin plan.

[0100] In an optional manner, inputting the intervention effect of medical staff, the real-time data feedback of patients, and the enteral nutrition complication risk warning system into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model further includes:

[0101] Real-time collection of patient data during enteral nutrition, including daily blood sugar, blood gas analysis, electrolytes, renal function, liver function, and inflammatory indicators;

[0102] Automatically collect information on the type, rate, dosage, and route of nutrient solution by connecting to an infusion pump, blood glucose meter, and monitor; obtain information from medical staff on adjusting nutrient solution formulas, infusion rates, medications, and body position management, as well as the corresponding reasons for intervention, intervention measures, and expected effects; and record the patient's subjective experience and the time, frequency, and severity of symptoms.

[0103] If the patient's gastric residual volume decreases after adjusting the nutrient solution infusion rate, the intervention is effective; if the patient's diarrhea symptoms are alleviated after taking probiotics, the intervention is effective; based on the assessment results of the enteral nutrition complication risk warning system, determine whether the patient's risk level has changed; if the patient's risk level decreases, the intervention is effective; otherwise, the intervention needs to be adjusted;

[0104] The expected effects of medical staff, the index data of patients and the enteral nutrition complication risk warning system are input into the dynamic risk prediction model, and the model parameters are automatically adjusted based on the new data.

[0105] In the present embodiment, for example, a patient developed hyperglycemia during enteral nutrition, and the medical staff adjusted the insulin dosage. Subsequently, the patient's blood sugar data, insulin dosage, nutrient solution type, infusion rate and other data are automatically collected. The medical staff records the reasons, measures and expected effects (such as lowering blood sugar) for adjusting the insulin dosage, and the patient has no other discomfort. After a period of observation, the patient's blood sugar is effectively controlled. The evaluation results of the enteral nutrition complication risk warning system show that the patient's metabolic complication risk is reduced. The above data are input into the dynamic risk prediction model, and the knowledge that "adjusting insulin dosage can effectively reduce the risk of hyperglycemia" is learned and the model parameters are automatically adjusted, so that the model pays more attention to the factor of insulin dosage when predicting the risks of similar patients.

[0106] According to the solution provided by the present invention, a multimodal data vector is generated based on the patient's clinical data, environmental data and behavioral data; wherein, the clinical data include blood glucose level, gastrointestinal function index, nutrient solution infusion speed and temperature, bowel movement diary and Braden scale, the environmental data include ward temperature and humidity and nutrient solution storage conditions, and the behavioral data include patient posture changes and activity frequency; the patient's multimodal data vector and individual feature information are input into a trained dynamic risk prediction model, and the probability of occurrence of gastrointestinal complications, mechanical complications, infectious complications and metabolic complications is output; wherein, the dynamic risk prediction model includes LSTM network layer, Transformer network layer, MLP network layer, Stacking network layer, Layer, cluster analysis layer and LIME network interpretation layer; the individual characteristic information includes age, underlying diseases and nutritional status; the cluster analysis layer is used to perform cluster analysis based on the probability of occurrence of each complication output by the Stacking network layer, and output risk grading results including four levels of low risk, medium risk, high risk and critical risk; based on the risk grading results, a personalized intervention plan for the patient is generated for reference by medical staff; the intervention effect of medical staff, real-time data feedback of patients and enteral nutrition complication risk warning system are input into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model; wherein, the enteral nutrition complication risk warning system includes a proposed entry pool, a first round of expert consultation and a second round of evaluation results. The present invention generates multimodal data vectors based on clinical data, environmental data and behavioral data, which can more comprehensively reflect the patient's health status and external environmental factors. The cluster analysis layer is used to divide the patient's risk level into four levels of low risk, medium risk, high risk and critical risk, which facilitates medical staff to quickly understand the patient's risk status. Based on risk stratification results, expert consensus, and clinical guidelines, differentiated enteral nutrition support plans and interventions are developed to provide patients with more personalized medical services. The LIME network interpretation layer helps medical staff better understand risk factors and implement more refined interventions. Factors associated with the risk of enteral nutrition complications are collected through literature review, clinical experience summary, and patient interviews. Expert consultation and evaluation ensure the reliability of the early warning system.

[0107] Figure 6 The schematic diagram of the framework of the enteral nutrition complication risk classification early warning management device according to an embodiment of the present invention is shown. The enteral nutrition complication risk classification early warning management device comprises:

[0108] A multimodal data vector generation module 610 is configured to generate a multimodal data vector based on the patient's clinical data, environmental data, and behavioral data; wherein the clinical data includes blood glucose level, gastrointestinal function indicators, nutrient solution infusion rate and temperature, bowel movement diary, and Braden scale; the environmental data includes ward temperature and humidity and nutrient solution storage conditions; and the behavioral data includes changes in patient position and activity frequency.

[0109] Dynamic risk prediction module 620 is used to input the patient's multimodal data vector and individual characteristic information into a trained dynamic risk prediction model, and output the probability of gastrointestinal complications, mechanical complications, infectious complications, and metabolic complications; wherein the dynamic risk prediction model includes an LSTM network layer, a Transformer network layer, an MLP network layer, a Stacking network layer, a clustering analysis layer, and a LIME network interpretation layer; the individual characteristic information includes age, underlying diseases, and nutritional status; the clustering analysis layer is used to perform cluster analysis based on the probability of occurrence of each complication output by the Stacking network layer, and output risk grading results including four levels of low risk, medium risk, high risk, and critical risk;

[0110] The personalized intervention plan generation module 630 is used to generate a personalized intervention plan for the patient based on the risk grading results for reference by medical staff; the intervention effect of medical staff, the real-time data feedback of the patient and the enteral nutrition complication risk warning system are input into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model; wherein, the enteral nutrition complication risk warning system includes a proposed item pool, a first round of expert inquiries and a second round of evaluation results.

[0111] Figure 7 The schematic diagram of the structure of the computing device embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0112] like Figure 7 As shown, the computing device may include: a processor (processor) 702 , a communications interface (Communications Interface) 704 , a memory (memory) 706 , and a communication bus 708 .

[0113] Processor 702, communication interface 704, and memory 706 communicate with each other via communication bus 708. Communication interface 704 is used to communicate with other devices, such as client devices or other server network elements. Processor 702 is used to execute program 710, which may specifically perform the steps described in the embodiment of the enteral nutrition complication risk grading and early warning management method.

[0114] Specifically, the program 710 may include program codes, which include computer operation instructions.

[0115] Processor 702 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in a computing device may be of the same type, such as one or more CPUs, or may be of different types, such as one or more CPUs and one or more ASICs.

[0116] The memory 706 is used to store the program 710. The memory 706 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0117] According to the solution provided by the present invention, a multimodal data vector is generated based on the patient's clinical data, environmental data and behavioral data; wherein, the clinical data include blood glucose level, gastrointestinal function index, nutrient solution infusion speed and temperature, bowel movement diary and Braden scale, the environmental data include ward temperature and humidity and nutrient solution storage conditions, and the behavioral data include patient posture changes and activity frequency; the patient's multimodal data vector and individual feature information are input into a trained dynamic risk prediction model, and the probability of occurrence of gastrointestinal complications, mechanical complications, infectious complications and metabolic complications is output; wherein, the dynamic risk prediction model includes LSTM network layer, Transformer network layer, MLP network layer, Stacking network layer, Layer, cluster analysis layer and LIME network interpretation layer; the individual characteristic information includes age, underlying diseases and nutritional status; the cluster analysis layer is used to perform cluster analysis based on the probability of occurrence of each complication output by the Stacking network layer, and output risk grading results including four levels of low risk, medium risk, high risk and critical risk; based on the risk grading results, a personalized intervention plan for the patient is generated for reference by medical staff; the intervention effect of medical staff, real-time data feedback of patients and enteral nutrition complication risk warning system are input into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model; wherein, the enteral nutrition complication risk warning system includes a proposed entry pool, a first round of expert consultation and a second round of evaluation results. The present invention generates multimodal data vectors based on clinical data, environmental data and behavioral data, which can more comprehensively reflect the patient's health status and external environmental factors. The cluster analysis layer is used to divide the patient's risk level into four levels of low risk, medium risk, high risk and critical risk, which facilitates medical staff to quickly understand the patient's risk status. Based on risk stratification results, expert consensus, and clinical guidelines, differentiated enteral nutrition support plans and interventions are developed to provide patients with more personalized medical services. The LIME network interpretation layer helps medical staff better understand risk factors and implement more refined interventions. Factors associated with the risk of enteral nutrition complications are collected through literature review, clinical experience summary, and patient interviews. Expert consultation and evaluation ensure the reliability of the early warning system.

[0118] Those skilled in the art will appreciate that modules in the devices of the embodiments may be adaptively modified and deployed in one or more devices different from the embodiments. Modules, units, or components in the embodiments may be combined into a single module, unit, or component, and furthermore, they may be divided into multiple submodules, subunits, or subcomponents. All features disclosed in this specification (including the accompanying claims, abstract, and drawings), as well as all processes or units of any method or device disclosed therein, may be combined in any combination, except where at least some of such features and / or processes or units are mutually exclusive. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that provides the same, equivalent, or similar purpose. Furthermore, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the present invention and to form different embodiments. For example, in the claims below, any of the claimed embodiments may be used in any combination. The present invention may be implemented using hardware comprising a number of different elements and using a suitably programmed computer. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. Unless otherwise specified, the steps in the above embodiments should not be understood as limiting the order of execution.

Claims

1. A risk classification and early warning management method for enteral nutrition complications, characterized in that: include: Generate a multimodal data vector based on the patient's clinical data, environmental data, and behavioral data; wherein the clinical data includes blood glucose level, gastrointestinal function indicators, nutrient solution infusion rate and temperature, bowel movement diary, and Braden scale; the environmental data includes ward temperature and humidity and nutrient solution storage conditions; and the behavioral data includes changes in patient position and activity frequency; The patient's multimodal data vector and individual characteristic information are input into a trained dynamic risk prediction model to output the probability of gastrointestinal complications, mechanical complications, infectious complications, and metabolic complications; wherein the dynamic risk prediction model includes an LSTM network layer, a Transformer network layer, an MLP network layer, a Stacking network layer, a clustering analysis layer, and a LIME network interpretation layer; the individual characteristic information includes age, underlying diseases, and nutritional status; the clustering analysis layer is used to perform cluster analysis based on the probability of occurrence of each complication output by the Stacking network layer, and output risk grading results including four levels of low risk, medium risk, high risk, and critical risk; A personalized intervention plan for the patient is generated based on the risk grading results for reference by medical staff; the intervention effect of medical staff, the real-time data feedback of the patient and the enteral nutrition complication risk warning system are input into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model; wherein, the enteral nutrition complication risk warning system includes a proposed item pool, a first round of expert inquiries and a second round of evaluation results.

2. The enteral nutrition complication risk classification and early warning management method according to claim 1, characterized in that: The LSTM network layer includes a blood glucose trend analysis unit, a gastrointestinal function monitoring unit, a nutrient solution infusion optimization unit, an environmental factor influence unit, a body position and activity analysis unit, a metabolic state prediction unit, an infection risk assessment unit, a time series feature extraction unit, a dynamic risk integration unit, and a long-term dependency capture unit; Wherein, the blood glucose trend analysis unit is used to analyze the trend of the patient's blood glucose level over time and predict its correlation with metabolic complications; The gastrointestinal function monitoring unit is used to analyze the changing patterns of gastric emptying time and intestinal motility frequency and predict the risk of gastrointestinal complications; The nutrient solution infusion optimization unit is used to monitor changes in nutrient solution infusion rate and temperature, and predict the risk of diarrhea and abdominal distension in gastrointestinal complications; the environmental factor impact unit is used to analyze the impact of ward temperature and humidity, and nutrient solution storage conditions on the patient's intestinal flora and infectious complications; The body position and activity analysis unit is used to analyze the effects of changes in the patient's body position and activity frequency on gastrointestinal function, and to predict the risk of gastrointestinal complications caused by improper body position or insufficient activity; The metabolic state prediction unit is used to integrate blood sugar level and nutrient solution infusion rate data and predict the risk of metabolic complications; The infection risk assessment unit analyzes the risk of infectious complications based on the environmental data and behavioral data; The time series feature extraction unit is used to integrate the time series features of clinical data, environmental data and behavioral data; The dynamic risk integration unit is used to generate comprehensive risk prediction results; The long-term dependency capture unit is used to capture the cumulative impact of a continuous increase in blood sugar levels on metabolic complications, or the potential threat of long-term abnormal temperature and humidity in the ward to infectious complications.

3. The enteral nutrition complication risk classification and early warning management method according to claim 1, characterized in that: The results of the first round of expert consultation and the second round of evaluation further include: Through literature review, clinical experience summary and patient interviews, we collected, screened and classified physiological factors, enteral nutrition factors, environmental factors and patient behavior factors related to the risk of enteral nutrition complications to form an initial pool of operation items; Invite a preset number of experts in the field of enteral nutrition, provide the initial pool of items to the experts, have the experts perform a first round of scoring on the initial pool of items, and screen the pool of items according to preset scoring criteria; The screened item pool is sent again to the same group of experts for a second round of evaluation. If the evaluation results of the second round are within the preset range of the evaluation results of the first round, the screened item pool reaches a stable state. The screened item pool is used as a component of the enteral nutrition complication risk warning system to assess the patient's risk factors.

4. The enteral nutrition complication risk classification and early warning management method according to claim 1, characterized in that: The Stacking network layer includes a basic model layer and a meta-learner layer; the LIME network interpretation layer includes a local perturbation module, a model prediction module, an interpreter module, a visualization module and a rule extraction module.

5. The enteral nutrition complication risk classification and early warning management method according to claim 1, characterized in that: Generating a personalized intervention plan for the patient according to the risk grading result further includes: Differentiated enteral nutrition support plans are developed for patients at different risk levels based on expert consensus, clinical guidelines, and individual patient conditions. For low-risk patients, formulas containing short peptides or medium-chain triglycerides are selected to promote absorption and reduce gastrointestinal burden. For medium-risk patients, digestion and absorption are assessed through abdominal auscultation and measurement of gastric residual volume, and the infusion rate and concentration of the nutrient solution are adjusted based on the monitoring results. For high-risk patients, nasojejunal or gastrojejunal tubes are used for infusion, bypassing the stomach to reduce the risk of aspiration. For patients at different risk of complications, a small dose and slowly increasing infusion rate are used to avoid overburdening the gastrointestinal tract. For patients at risk of mechanical complications, the catheter position is regularly checked to ensure catheter patency. For patients at risk of metabolic complications, the nutrient solution formula and infusion rate are adjusted, and insulin medication is used to control blood sugar, with fine-tuning performed based on the rules provided by the LIME network interpretation layer. The intervention plan is dynamically adjusted according to the patient's actual situation and intervention effect for reference by medical staff; if the patient has diarrhea, the concentration of the nutrient solution is reduced, the infusion rate is adjusted, or the nutrient solution formula is changed; if the patient's blood sugar continues to rise, the insulin dosage is adjusted or the nutrient solution type is changed.

6. The enteral nutrition complication risk classification and early warning management method according to claim 1, characterized in that: The training process of the dynamic risk prediction model further includes: Divide the historical multimodal data vector into training set, validation set and test set; The LSTM network layer is trained using the training set to learn long-term dependencies in time series data; the output of the LSTM network layer is used as the input of the Transformer network layer to learn the global relationship between the data; the output of the Transformer network layer and individual feature information are input into the MLP network layer for feature fusion and nonlinear transformation; the output of the MLP network layer is input into the Stacking network layer to perform ensemble learning using multiple basic models; and the number of hidden units, learning rate, and dropout rate of the LSTM network layer and the Transformer network layer are adjusted using the validation set; After the dynamic risk prediction model is trained, the LIME network interpretation layer is used to interpret the prediction results. By making local perturbations near the prediction point and observing the changes in the prediction results, the importance of each feature to the prediction results is inferred.

7. The enteral nutrition complication risk classification and early warning management method according to claim 1, characterized in that: Inputting the intervention effect of medical staff, the real-time data feedback of patients and the enteral nutrition complication risk warning system into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model further includes: Real-time collection of patient data during enteral nutrition, including daily blood sugar, blood gas analysis, electrolytes, renal function, liver function, and inflammatory indicators; Automatically collect information on the type, rate, dosage, and route of nutrient solution by connecting to an infusion pump, blood glucose meter, and monitor; obtain information from medical staff on adjusting nutrient solution formulas, infusion rates, medications, and body position management, as well as the corresponding reasons for intervention, intervention measures, and expected effects; and record the patient's subjective experience and the time, frequency, and severity of symptoms. If the patient's gastric residual volume decreases after adjusting the nutrient solution infusion rate, the intervention is effective; if the patient's diarrhea symptoms are alleviated after taking probiotics, the intervention is effective; based on the assessment results of the enteral nutrition complication risk warning system, determine whether the patient's risk level has changed; if the patient's risk level decreases, the intervention is effective; otherwise, the intervention needs to be adjusted; The expected effects of medical staff, the index data of patients and the enteral nutrition complication risk warning system are input into the dynamic risk prediction model, and the model parameters are automatically adjusted based on the new data.

8. The enteral nutrition complication risk classification and early warning management method according to claim 1, characterized in that: The clustering formula of the cluster analysis layer is: ； in, As the core point; The probability vector of each complication of the patient output by the Stacking network layer; For data points Distance less than or equal to The set of all data points; is the neighborhood radius threshold; is the minimum number of neighbor points; For data points At a radius of within the scope the average density of neighbors; is the neighbor parameter; is the density threshold; For data points The distance to the cluster center Centroid; is the distance coefficient; For data points The maximum distance to the cluster center to which it belongs.

9. A device for early warning and management of risk classification of enteral nutrition complications, characterized in that: include: a multimodal data vector generation module, configured to generate a multimodal data vector based on the patient's clinical data, environmental data, and behavioral data; wherein the clinical data includes blood glucose level, gastrointestinal function indicators, nutrient solution infusion rate and temperature, bowel movement diary, and Braden scale; the environmental data includes ward temperature and humidity and nutrient solution storage conditions; and the behavioral data includes changes in patient position and activity frequency; A dynamic risk prediction module is used to input the patient's multimodal data vector and individual characteristic information into a trained dynamic risk prediction model, and output the probability of gastrointestinal complications, mechanical complications, infectious complications, and metabolic complications; wherein the dynamic risk prediction model includes an LSTM network layer, a Transformer network layer, an MLP network layer, a Stacking network layer, a clustering analysis layer, and a LIME network interpretation layer; the individual characteristic information includes age, underlying diseases, and nutritional status; the clustering analysis layer is used to perform cluster analysis based on the probability of occurrence of each complication output by the Stacking network layer, and output risk grading results including four levels of low risk, medium risk, high risk, and critical risk; A personalized intervention plan generation module is used to generate a personalized intervention plan for the patient based on the risk grading results for reference by medical staff; the intervention effect of medical staff, the real-time data feedback of patients and the enteral nutrition complication risk warning system are input into the dynamic risk prediction model to adjust the model parameters of the dynamic risk prediction model; wherein, the enteral nutrition complication risk warning system includes a proposed item pool, the first round of expert inquiries and the second round of evaluation results.

10. A computing device comprising: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned enteral nutrition complication risk classification warning management method.

Images