Enteral nutritional complication risk assessment system and early warning method
By acquiring and analyzing multimodal risk factors in enteral nutrition patients, and utilizing DTW and Bi-LSTM models, the problem of distorted assessment results caused by individual differences in traditional assessment models was solved, achieving more accurate risk assessment and early warning of complications.
Patent Information
- Application Number
- CN202511660799.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-13
- Publication Date
- 2025-12-23
AI Technical Summary
In existing technologies, enteral nutrition complication assessment models based on population statistical patterns lack characterization of individual patient-specific mechanisms, leading to distorted assessment results and insufficient accuracy.
By acquiring multimodal risk factors over time for each patient with enteral nutrition complications, and using the Dynamic Time Warping (DTW) algorithm and Bi-LSTM model, the fluctuation characteristics of individual patient risk factors are analyzed to determine the degree of personalization and causal weights, thereby constructing a personalized complication risk assessment system.
It improves the accuracy of risk assessment for enteral nutrition complications, identifies the core factors most worthy of monitoring for individual patients, reduces misjudgments and underreporting, and achieves precise early warning.
Smart Images
Figure CN121191771A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of health risk assessment, in particular to an enteral nutrition complication risk assessment system and early warning method. BACKGROUND
[0002] Enteral nutrition is one of the most commonly used nutritional support methods in clinical practice, and is widely used in critically ill patients, postoperative rehabilitation and patients who cannot eat orally. Enteral nutrition can better maintain intestinal function, improve nutritional status and reduce the risk of infection. However, in the actual application process, enteral nutrition is often accompanied by the occurrence of various complications, such as gastrointestinal intolerance (abdominal distension, diarrhea, constipation, vomiting), metabolic disorders and respiratory complications caused by aspiration. At present, the probability of a certain patient developing a certain complication is usually predicted by analyzing and integrating multidimensional data of a large number of patients to construct a mathematical model, so as to realize the assessment and early warning of enteral nutrition complications.
[0003] In the prior art, in clinical practice, patients have significant individual differences in underlying diseases, intestinal flora composition, metabolic levels and tolerance to enteral nutrition, which can lead to completely different complication manifestations in different patients with the same nutritional method and dose. Traditional assessment models are mostly based on group statistical rules and can only establish a correlation between risk factors and complications for prediction, but lack of characterization of patient-specific mechanisms. When facing a patient group with large individual differences, the traditional method is prone to distorted evaluation results, that is, the accuracy of risk assessment is insufficient. SUMMARY
[0004] In order to solve the technical problems of correlation prediction based on group statistical rules in the related art, which is prone to distorted evaluation results and insufficient accuracy of risk assessment, the present application provides an enteral nutrition complication risk assessment system and early warning method, and the technical solutions adopted are as follows: The present application provides an enteral nutrition complication risk assessment and early warning method, which comprises the following steps: Obtaining the time sequence of multiple modal risk factors of each enteral nutrition complication patient; determining the fluctuation coefficient of each type of risk factor according to the numerical fluctuation of the risk factor of a single mode of any two patients; For each type of risk factor of each patient, the fluctuation similarity and fluctuation coefficient of each type of risk factor of all patients are combined to determine the individualization degree; the key degree is determined according to the time sequence change amplitude of the same type of risk factor; the special index of each patient's each type of risk factor for complication assessment is determined by combining the key degree and the individualization degree; determine the influence degree of the two risk factors according to the similarity of the numerical fluctuation of each patient in the two risk factors; obtain the risk factor chain of each complication; determine the chain reaction index according to the probability of each type of risk factor leading to the occurrence of the complication on the chain, and the influence degree of each type of risk factor on other risk factors on the chain; determine the causal weight of each type of risk factor leading to the complication in combination with the special index and the chain reaction index of each type of risk factor of all patients, and conduct risk assessment and early warning in combination with the causal weight of all risk factors.
[0005] Further, the determination of the fluctuation coefficient of each type of risk factor according to the numerical fluctuation of the risk factor of a single mode of any two patients comprises: sequentially arrange each risk factor of each patient to obtain a risk factor sequence; calculate the DTW matching distance of the risk factor sequence of the same risk factor of any two patients, and normalize the DTW matching distance as the fluctuation difference value of the same risk factor of the two patients; statistically analyze all combinations of any two patients, and take the mean value of the fluctuation difference values of all combinations as the fluctuation coefficient of the corresponding type of risk factor.
[0006] Further, the determination of the individualization degree in combination with the fluctuation similarity and the fluctuation coefficient of each type of risk factor of all patients comprises: calculate the mean value of the risk factor of all patients at the same analysis time to form a mean value sequence, wherein the analysis time is the time information obtained by aligning each type of risk factor of different patients in time according to the time stamp; calculate the DTW matching distance of the risk factor sequence of each patient and the mean value sequence, and normalize the DTW matching distance as the overall difference index; normalize the difference value between the overall difference index and the fluctuation coefficient as the individualization degree of the corresponding type of risk factor of the corresponding patient.
[0007] Further, the determination of the criticality degree according to the time sequence change amplitude of the same type of risk factor comprises: calculate the numerical standard deviation of the same type of risk factor of the same patient, and normalize the numerical standard deviation as the criticality degree of the corresponding type of risk factor of the corresponding patient.
[0008] Further, the determination of the special index of each type of risk factor of each patient for the complication assessment in combination with the criticality degree and the individualization degree comprises: calculate the product value of the criticality degree and the individualization degree, and normalize the product value as the special index.
[0009] Further, the determining the influence degree of the two risk factors according to the similarity of the numerical fluctuation of each patient in any two risk factors comprises: determining the cosine similarity between each patient in any two risk factors respectively composed of a sequence, and normalizing the correlation of the two risk factors as the correlation; calculating the average correlation of any two risk factors as the total correlation coefficient; normalizing the difference between the correlation of the two risk factors and the total correlation coefficient as the influence degree between the two risk factors.
[0010] Further, the determining the chain reaction index according to the probability of each type of risk factor on the chain causing complications and the influence degree of each type of risk factor on other risk factors on the chain comprises: based on the correlation between the risk factor and the complication obtained by the traditional regression model, the regression coefficient is mapped by sigmoid to obtain the complication probability of each risk factor; calculating the average of the influence degree of any risk factor and all other risk factors on the chain as the influence analysis index; normalizing the product value of the influence analysis index and the probability as the chain reaction index.
[0011] Further, the determining the causal weight of each type of risk factor causing complications by combining the specificity index and the chain reaction index of each type of risk factor of all patients comprises: calculating the product value of the specificity index and the chain reaction index, and normalizing it as the causal weight of the risk factor causing complications.
[0012] Further, the risk assessment and early warning by combining the causal weight of all risk factors comprises: using the causal weight as the weight of each patient in the Bi-LSTM model, constructing a Bi-LSTM model to extract the time sequence features most decisive to the risk of complications, and realizing risk assessment and early warning.
[0013] On the other hand, an enteral nutrition complication risk assessment system is also provided, which comprises a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of the method according to any one of the preceding embodiments when executing the computer program.
[0014] The present application has the following advantages: The embodiment of the present application can avoid the deviation caused by simply relying on the average value of the population by comparing the statistical rules of a large number of population samples through acquiring the multi-modal risk factors of each enteral nutrition complication patient in time sequence, and can capture the abnormal fluctuation mode specific to different patients. Then, the individualization degree of each patient for the numerical fluctuation of each risk factor of a single mode is determined according to the numerical fluctuation and fluctuation similarity of the risk factor of the patient, the individualization degree can accurately represent the individualization characteristics of the numerical fluctuation of the corresponding risk factor of the patient, the special index of each patient for each type of risk factor for complication evaluation is determined by combining the key degree and the individualization degree, the difference of different factors in different patients can be determined, and the specific contribution of the risk factor is highlighted. The risk factors often do not act independently, but there is a causal transmission relationship. By analyzing the correlation between these factors, the formation mechanism of the complication can be revealed, the key chain can be avoided by analyzing a single factor in isolation, the causal weight of the risk factor is obtained by calculating the influence strength of the risk factor in the causal chain, the core factor most worthy of monitoring for the patient individual is identified, and the accuracy of the complication risk assessment is improved. BRIEF DESCRIPTION OF DRAWINGS
[0015] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.
[0016] Figure 1 A flow chart of an enteral nutrition complication risk assessment and early warning method provided by an embodiment of the present application; Figure 2 A DTW matching schematic diagram provided by an embodiment of the present application. DETAILED DESCRIPTION
[0017] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of the enteral nutrition complication risk assessment system and early warning method according to the present application are described in detail as follows by combining with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.
[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.
[0019] Enteral nutrition is one of the most commonly used nutritional support methods in clinical practice, and is widely used in critically ill patients, postoperative rehabilitation and patients who cannot take food orally. Enteral nutrition can better maintain intestinal function, improve nutritional status and reduce the risk of infection. However, in the actual application process, enteral nutrition is often accompanied by a variety of complications, such as gastrointestinal intolerance (abdominal distension, diarrhea, constipation, vomiting), metabolic disorders and respiratory complications caused by aspiration. At present, based on a large number of multi-dimensional data of patients, a mathematical model is constructed by analysis and integration to predict the probability of a certain patient developing a certain complication, so as to realize the evaluation and early warning of enteral nutrition complications.
[0020] In the prior art, in clinical practice, there are significant individual differences in patients in terms of underlying diseases, intestinal flora composition, metabolic level and tolerance to enteral nutrition, resulting in that the same nutritional method and dose may cause completely different complication manifestations in different patients. The traditional evaluation model is mostly based on group statistical rules, and can only establish the correlation between risk factors and complications for prediction, but lacks the characterization of patient-specific mechanisms. When facing a patient group with large individual differences, the traditional method is prone to distorted evaluation results, such as excessive warning of risks for some patients, increasing the burden of intervention; for other patients, it may be missed, delaying the best treatment opportunity.
[0021] Based on the above technical problems, the embodiments of the present application make specific analysis.
[0022] The specific scheme of the enteral nutrition complication risk assessment and early warning method provided by the present application will be specifically described below in combination with the drawings.
[0023] Please refer to Figure 1 which shows a flow chart of an enteral nutrition complication risk assessment and early warning method provided by an embodiment of the present application, and the method comprises: S101: acquiring multi-modal risk factors of each enteral nutrition complication patient in time sequence; and determining the fluctuation coefficient of each type of risk factor according to the numerical fluctuation of the risk factor of a single mode of any two patients.
[0024] In the enteral nutrition complication risk assessment, multi-modal data needs to be covered: 1. Clinical indicators: blood routine, CRP, PCT, biochemistry, electrolyte, blood glucose; 2. Disease signs: heart rate, blood pressure, body temperature; 3. Enteral nutrition operation: nutritional preparation type, concentration, infusion volume and mode; 4. Nursing records and symptoms: gastric residual volume, vomiting / reflux, diarrhea frequency, constipation, abdominal distension; 5. Microbiome and metabolome: feces, fecal metabolites, serum short-chain fatty acids (SCFA), bile acid spectrum.
[0025] It should be noted that in the embodiments of the present application, the data are quantified, and for text data, vectorization is performed through semantic analysis to achieve quantization effect, and after quantization, multi-modal data are obtained, which facilitates subsequent data processing.
[0026] In the embodiments of the present application, the multi-modal data need to be preprocessed, the multi-modal data are aligned according to timestamps, and the time sequence is unified; interpolation is used for short-term physiological monitoring; batch correction and abnormal value detection are performed; and the above multi-modal data are normalized.
[0027] The same type of risk factor, for example, the numerical change of heart rate, can represent the fluctuation degree, and the fluctuation degree of different patients will differ, and the fluctuation coefficient is used to realize the comparison effect of the difference.
[0028] According to the numerical fluctuation of the risk factor of a single mode of any two patients, the fluctuation coefficient of each type of risk factor is determined, including: arranging each risk factor of each patient in time sequence to obtain a risk factor sequence; calculating the DTW matching distance of the risk factor sequence of any two patients for the same risk factor, and normalizing the DTW matching distance as the fluctuation difference value of the same risk factor of the two patients; and statistically combining all combinations, and taking the mean value of the fluctuation difference values of all combinations as the fluctuation coefficient of the corresponding type of risk factor.
[0029] Referring to Figure 2 , Figure 2 The DTW matching schematic diagram provided by an embodiment of the present application is used to arrange the risk factors in time sequence to obtain a fitting fluctuation sequence at different sampling times, which is embodied as two fluctuation curves in Figure 2 , and the fitting fluctuations of the same risk factor of any two patients are analyzed.
[0030] Among them, the time change sequence of the risk factor is constructed, and then the fluctuation inconsistency analysis is performed based on the fluctuations of any two patients, and the fluctuation inconsistency analysis in the embodiments of the present application is the DTW matching based on the dynamic time warping algorithm, and the DTW matching distance obtained by the DTW matching is normalized as the fluctuation difference value.
[0031] The greater the numerical value of the DTW matching distance is, the more inconsistent the fluctuations between the risk factor sequences are, the higher the difference is, that is, the greater the numerical value of the fluctuation difference is, and then the combinations of every two patients are counted, and the mean value of the fluctuation difference values of all combinations is taken as the fluctuation coefficient of the corresponding type of risk factor. The greater the numerical value of the fluctuation coefficient is, the greater the dynamic change trend of the risk factor in the patient population is.
[0032] S102: For each patient, each type of risk factor, combined with the fluctuation similarity and the fluctuation coefficient of each type of risk factor of all patients, determine the degree of individualization; according to the time sequence change range of the same type of risk factor, determine the key degree; combine the key degree and the individualization degree, determine the special index of each patient each type of risk factor to the complication evaluation.
[0033] In the process of enteral nutrition complication risk assessment, patient's own factors, such as gastrointestinal dysfunction, basic disease influence or low immunity, etc., are easy to cause the risk factor change to deviate obviously from the group change, and this difference prompts that the risk factor has special significance for the complication occurrence of the patient, and should not be judged only by the average rule of the group, otherwise the individual risk is easy to be ignored, thus leading to risk underestimation or misjudgment. Therefore, the difference analysis can highlight the patient-specific (special) risk characteristics, and provide reliable basis for individualized risk assessment and accurate early warning.
[0034] Generally, the individualized risk factor of the patient to be evaluated for enteral nutrition complication is obtained by the dynamic change of the risk factor and the difference from the group, however, if the dynamic change trend of the risk factor that may cause complications exists certain difference in the same group of patients, at this time, the difference of the patient to be evaluated cannot highlight its individualized characteristics. Therefore, by confirming the fluctuation coefficient of the risk factor among the same group of patients, when the dynamic change similarity of the risk factor between any two patients in the same group of patients exists difference, it indicates that the dynamic change trend of the risk factor in the same group of patients exists fluctuation.
[0035] In order to further analyze the fluctuation of the change trend of each patient in the group, it is also necessary to combine the fluctuation similarity of each type of risk factor of all patients.
[0036] Based on this, in the embodiment of the application, the degree of individualization is determined by combining the fluctuation similarity and the fluctuation coefficient of each type of risk factor of all patients, including: calculating the numerical mean of the risk factor of all patients at the same analysis time, to form a mean sequence, wherein the analysis time is the time information obtained by aligning each type of risk factor of different patients according to the time stamp and unifying the time sequence; calculating the DTW matching distance of the risk factor sequence of each patient and the mean sequence, and normalizing the whole difference index; the difference between the whole difference index and the fluctuation coefficient is normalized as the individualization degree of the corresponding patient for the corresponding type of risk factor.
[0037] In the embodiment of the application, the step of aligning each type of risk factor of different patients by time stamp is a data preprocessing step, and the time effectiveness information is unified by preprocessing. In the embodiment of the application, the first day of each risk detection is taken as the initial time, and the subsequent time stamp alignment process is carried out.
[0038] Wherein, the mean sequence represents the numerical fluctuation of the whole, and the DTW matching distance between the risk factor sequence of each patient and the mean sequence is calculated to obtain the whole difference index, and the greater the value is, the greater the fluctuation difference between the patient and the whole is, that is, the patient has a higher degree of individualization in the corresponding risk factor, and similarly, the smaller the value of the fluctuation coefficient is, the more stable the overall waveform of all patients in the corresponding risk factor is, and the higher the reliability of the obtained individualization degree is, based on the above characteristics, the embodiment of the application calculates the difference between the whole difference index and the fluctuation coefficient, and normalizes the difference as the individualization degree of the corresponding patient to the corresponding risk factor of the same type.
[0039] When the change range of the risk factor of the patient in the continuous detection is relatively large, the risk factor is more critical. Therefore, in the embodiment of the application, the criticality is determined according to the time sequence change range of the risk factor of the same type, including: calculating the numerical standard deviation of the risk factor of the same type of the same patient, and normalizing the numerical standard deviation as the criticality of the corresponding patient to the corresponding risk factor of the same type.
[0040] Wherein, the greater the numerical standard deviation is, the greater the change range of the risk factor of the patient in the continuous detection is, and the higher the criticality is.
[0041] By combining the criticality and the individualization degree of the risk factor, the speciality of each risk factor in the specific patient for the complication evaluation can be quantified, so that the most influential and most concerned factors can be identified.
[0042] Further, in some embodiments of the application, the criticality and the individualization degree are combined to determine the speciality index of each patient to each type of risk factor for the complication evaluation, including: calculating the product value of the criticality and the individualization degree, and normalizing the product value as the speciality index.
[0043] For any patient, the greater the numerical value of the speciality index of the corresponding risk factor is, the greater the influence of the risk factor on the patient is, and the more special attention needs to be paid.
[0044] S103: determining the influence degree of two risk factors according to the numerical fluctuation similarity of each patient in the two risk factors; obtaining the risk factor chain of each complication; and determining the chain reaction index according to the occurrence probability of each type of risk factor leading to the complication and the influence degree of each type of risk factor on other risk factors in the chain.
[0045] In the embodiment of the application, the higher the similarity of the fluctuation of different risk factors is, the higher the mutual influence degree is.
[0046] Therefore, in the embodiment of the present application, the influence degree of the two risk factors is determined according to the similarity of the numerical fluctuation of each patient in any two risk factors, including: determining the cosine similarity between the sequence composed of any two risk factors respectively, and normalizing the cosine similarity as the correlation of the two risk factors; calculating the average correlation of any two risk factors as the total correlation coefficient; and normalizing the difference between the correlation of the two risk factors and the total correlation coefficient as the influence degree between the two risk factors.
[0047] The cosine similarity between the sequence composed of any two risk factors respectively is the correlation, and the higher the value is, the greater the similarity of the changes of the two risk factors is. By analyzing the correlation of all combinations of the two risk factors, the total correlation coefficient representing the overall correlation is determined, so that the difference between the single two risk factors and the whole is normalized as the influence degree.
[0048] The greater the numerical value of the influence degree is, the greater the dominant influence of the corresponding risk factor on the other risk factor is, which facilitates the subsequent chain analysis of the cause-effect transmission chain.
[0049] In order to more accurately identify the key driving link that really leads to complications, the originally scattered correlation information is organized into a cause-effect transmission chain. In the embodiment of the present application, the risk factor chain of each complication is obtained through a cause-effect diagram model.
[0050] Since the same risk factor of different patients has different positions and weights in the chain (some people are sensitive to butyric acid, and some people are more sensitive to electrolytes), the chain weight can reflect individual differences. At the same time, through multivariate chain analysis, it is judged whether it is a real pathological transmission, which can filter incidental correlations, reduce over-intervention on low-risk patients, and also reduce the risk of missing reports on high-risk patients.
[0051] There is a chain reaction in enteral nutrition complications (such as imbalance of flora → metabolic disorder → mucosal damage → aspiration and other complications), in order to analyze the importance of risk factors in the chain leading to complications, it is necessary to analyze the correlation between risk factors, because some factors may not act alone, but form a chain reaction (influence degree) through mutual influence with other related factors, thereby amplifying or weakening the overall impact on complications.
[0052] When the change of a certain risk factor leads to a greater probability of the occurrence of complications, and the influence degree of the other risk factors in the chain leading to complications is also greater, it means that the risk factor is more important in the risk assessment of complications, and its chain reaction degree is higher. The chain reaction index of each risk factor of each patient is obtained, which represents the chain reaction degree.
[0053] Further, in some embodiments of the present application, according to the probability of each type of risk factor on the chain causing the occurrence of complications, and the influence degree of each type of risk factor on other risk factors on the chain, the chain reaction index is determined, including: based on the correlation between the risk factor and the complication obtained by the traditional regression model, the regression coefficient is mapped by sigmoid to obtain the complication occurrence probability of each risk factor; the mean value of the influence degree of any risk factor and all other risk factors on the chain is calculated as the influence analysis index; and the product value of the influence analysis index and the occurrence probability is normalized as the chain reaction index.
[0054] It should be noted that the correlation obtained by using the traditional regression model, and the regression coefficient is mapped by sigmoid to obtain the complication occurrence probability of each risk factor, which is prior art, and will not be further limited and described.
[0055] Then, the mean value of the influence degree of any risk factor and all other risk factors on the chain is calculated as the influence analysis index; the greater the value of the influence analysis index, the greater the influence degree of the risk factor on all other risk factors on the chain, that is, the risk factor itself has greater reference in the chain reaction.
[0056] In summary, in the embodiments of the present application, the product value of the influence analysis index and the occurrence probability is normalized as the chain reaction index.
[0057] The greater the value of the chain reaction index, the more important the influence of the risk factor on the overall complication risk factor chain.
[0058] S104: Determine the causal weight of each risk factor causing complications by combining the specificity index and the chain reaction index of each type of risk factor of all patients, and perform risk assessment and early warning by combining the causal weights of all risk factors.
[0059] In the risk assessment of enteral nutrition complications, the introduction of causal weight can quantify the strength of different risk factors causing complications through chain reaction. The risk factor triggers the complication through a series of intermediate links, and the size of the effect can be mapped as a causal weight matrix, thereby representing the degree of action between risk factors. Therefore, higher causal weight is given to the key risk factors highly related to the patient, so that the model is more sensitive to the changes of these risk factors, thereby improving the accuracy of individualized risk assessment and the pertinence of early warning.
[0060] Further, in some embodiments of the present application, the causal weight of each risk factor causing complications is determined by combining the specificity index and the chain reaction index of each type of risk factor of all patients, including: calculating the product value of the specificity index and the chain reaction index, and normalizing as the causal weight of the risk factor causing complications.
[0061] The risk factors of each patient not only have individual differences when triggering complications, but also can affect other related factors through a chain reaction, thereby indirectly or superimposedly increasing the risk of complications. By comprehensively considering the particularity and chain reaction degree of each risk factor, the causal weight of the risk factor to the complication can be calculated, the real contribution of each factor in the individualized risk assessment can be quantified, and more accurate risk prediction can be realized.
[0062] Therefore, the embodiment of the application directly calculates the product value of the particularity index and the chain reaction index, and normalizes the causal weight of the risk factor to the complication.
[0063] The causal weight is used as the weight of each risk factor of each patient to the complication in the Bi-LSTM model, a Bi-LSTM model is constructed to extract the most decisive time sequence features of the complication risk, and risk assessment and early warning are realized.
[0064] The Bi-LSTM model is used to model the multi-dimensional risk factor time sequence data of the patient, the bidirectional structure of the Bi-LSTM model can simultaneously capture the dynamic dependence relationship of the previous and subsequent time, and the attention mechanism can give a higher weight to the risk factor on this basis, so as to extract the most decisive time sequence features of the complication risk. Then, the extracted risk features are embedded into a causal graph model to construct a chain reaction causal network between different risk factors, and the step-by-step conduction path from the initial change to the occurrence of the complication is revealed.
[0065] Under this framework, the causal weight of the action strength between each factor in the network is further quantified to obtain the direct and indirect contribution of each risk factor to the triggering of the complication. Finally, the causal weight is combined with the individual dynamic characteristics of the patient to generate individualized risk assessment and early warning results, which can not only ensure the accuracy of the assessment, but also realize sensitive monitoring and early warning of high-risk factors.
[0066] The way of risk assessment and early warning based on the causal weight of the risk factor is a known technical solution in the art, and will not be described here.
[0067] The embodiments of the present application can avoid the deviation caused by simply relying on the population average by acquiring the multi-modal risk factors of each enteral nutrition complication patient in time sequence, comparing the statistical rules of a large number of population samples, and capturing the abnormal fluctuation mode specific to different patients. Then, the individualization degree of each patient for the numerical fluctuation of the risk factor of a single mode and the fluctuation similarity are determined, the individualization degree can accurately represent the individualization characteristics of the numerical fluctuation of the corresponding risk factor of the patient, the speciality index of each patient for each type of risk factor for the complication evaluation is determined by combining the key degree and the individualization degree, the difference of different factors in different patients can be determined, and the specific contribution of the risk factor is highlighted. The risk factors often do not act independently, but there is a causal transmission relationship. By analyzing the correlation between these factors, the formation mechanism of the complication can be revealed, the key chain can be avoided by analyzing a single factor in isolation, the influence strength of the risk factor in the causal chain is calculated to obtain the causal weight of the risk factor, the core factor most worthy of monitoring for the patient individual is identified, and the accuracy of the complication risk evaluation is improved.
[0068] In another aspect, an enteral nutrition complication risk assessment system is also provided, which comprises a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of the method according to any one of the preceding aspects when executing the computer program.
[0069] It should be noted that the above-mentioned embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or can be advantageous.
[0070] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment mainly describes the difference from other embodiments.
Claims
1. A method for risk assessment and early warning of enteral nutrition complications, characterized in that, The method includes: Obtain multimodal risk factors for each patient with enteral nutrition complications over time; determine the fluctuation coefficient of each type of risk factor based on the numerical fluctuation of risk factors for any two patients with a single modality. For each patient and each risk factor, the degree of personalization is determined by combining the similarity of fluctuations and the fluctuation coefficient of each risk factor across all patients; the degree of criticality is determined based on the temporal variation of the same risk factor; and the specific indicators for each patient and each risk factor in complication assessment are determined by combining the degree of criticality and the degree of personalization. Based on the similarity of the numerical fluctuations of each patient for any two risk factors, determine the degree of influence of the two risk factors; obtain the risk factor chain for each complication; determine the chain reaction index based on the probability of each type of risk factor in the chain causing the complication, and the degree of influence of each type of risk factor on other risk factors in the chain. By combining the specific indicators and chain reaction indicators of each type of risk factor for all patients, the causal weight of each risk factor leading to complications is determined, and risk assessment and early warning are carried out by combining the causal weights of all risk factors.
2. The method for risk assessment and early warning of enteral nutrition complications as described in claim 1, characterized in that, The determination of the volatility coefficient for each type of risk factor based on the numerical volatility of risk factors for any two patients with respect to a single modality includes: Each patient and each risk factor is arranged chronologically to obtain a risk factor sequence; Calculate the DTW matching distance for the risk factor sequences of any two patients for the same risk factor, and normalize the DTW matching distance as the fluctuation difference value between the two patients for the same risk factor. Statistically analyze all combinations of any two patients and use the mean of the volatility differences among all combinations as the volatility coefficient of the corresponding risk factor class.
3. The method for risk assessment and early warning of enteral nutrition complications as described in claim 2, characterized in that, The determination of individualization, which combines the volatility similarity and volatility coefficient of each risk factor for all patients, includes: Calculate the mean values of risk factors for all patients at the same analysis time and form a mean sequence. The analysis time is the time information obtained by aligning each type of risk factor for different patients according to the timestamp and unifying the time sequence. The DTW matching distance between the risk factor sequence and the mean sequence for each patient was calculated and normalized as an overall difference indicator. The difference between the overall difference index and the volatility coefficient is normalized to represent the degree of personalization of the corresponding risk factor for the corresponding patient.
4. The method for risk assessment and early warning of enteral nutrition complications as described in claim 1, characterized in that, The determination of criticality based on the temporal variation magnitude of the same type of risk factors includes: Calculate the numerical standard deviation of the same risk factor for the same patient, and normalize it to represent the criticality of the corresponding risk factor for the corresponding patient.
5. The method for risk assessment and early warning of enteral nutrition complications as described in claim 1, characterized in that, The combination of criticality and individualization determines specific indicators for each patient and each risk factor in assessing complications, including: The product of criticality and personalization is calculated and normalized as a specificity indicator.
6. The method for risk assessment and early warning of enteral nutrition complications as described in claim 1, characterized in that, The determination of the degree of influence of two risk factors based on the similarity of the numerical fluctuations of each patient across any two risk factors includes: Determine the cosine similarity between the sequences composed of any two risk factors for each patient, and normalize the results to represent the correlation between the two risk factors. Calculate the mean correlation between any two risk factors and use it as the overall correlation coefficient; The difference between the correlation coefficient of the two risk factors and the total correlation coefficient is normalized and used as the degree of influence between the two risk factors.
7. The method for risk assessment and early warning of enteral nutrition complications as described in claim 1, characterized in that, The chain reaction index is determined based on the probability of complications caused by each risk factor in the chain, and the degree of influence of each risk factor on other risk factors in the chain, including: Based on the association between risk factors and complications obtained from the traditional regression model, the regression coefficients are mapped using a sigmoid method to obtain the probability of complication for each risk factor. Calculate the average impact of any risk factor on all other risk factors in the chain, and use it as an impact analysis indicator; The product of the impact analysis index and the probability of occurrence is normalized and used as a chain reaction index.
8. The method for risk assessment and early warning of enteral nutrition complications as described in claim 1, characterized in that, The method combines specific indicators and chain reaction indicators for each type of risk factor in all patients to determine the causal weight of each risk factor leading to complications, including: The product of the specificity index and the chain reaction index is calculated and normalized to serve as the causal weight of the risk factor leading to complications.
9. The method for risk assessment and early warning of enteral nutrition complications as described in claim 1, characterized in that, The risk assessment and early warning system, which combines the causal weights of all risk factors, includes: By using causal weights as the weights of each risk factor leading to complications in each patient in the Bi-LSTM model, a Bi-LSTM model is constructed to extract the time-series features that are most decisive for complication risk, thereby achieving risk assessment and early warning.
10. A risk assessment system for enteral nutrition complications, the system comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 9.