A system and method for predicting early risk of sepsis

Abstract

The present invention relates to the technical field of health data prediction, and specifically to a prediction system and method for the early risk of sepsis. The method obtains the performance correlation of each sign detection type by analyzing the differences in clinical manifestations between sepsis patients and non-sepsis patients, as well as the correlation between multiple sign detection types and clinical manifestations; combines the differentiation strength of sepsis patients and non-sepsis patients in different sign detection types, and the numerical distribution of sepsis patients in sign detection types, to obtain the importance of each sign detection type, and obtains a sepsis risk model based on the importance and numerical value of the sign detection type. The present invention reflects the importance through the specific differentiation degree of the patient's sign detection type, the degree of characterization stability, and the characterization of clinical manifestations, and obtains a more accurate model through the importance, so that the risk prediction accuracy is more accurate and reliable.

Description

Technical Field

[0001] The present invention relates to the technical field of health data prediction, and in particular to a system and method for predicting the early risk of sepsis. Background Art

[0002] Sepsis is a general term for surgical infections with systemic inflammatory response manifestations, such as changes in body temperature, respiration, and circulation. It is the damage to body tissues caused by the absorption of endotoxins, exotoxins produced by pathogens and the various inflammatory mediators mediated by them. The current status of early risk prediction of sepsis mainly relies on clinical monitoring data and biomarker analysis. For example, physical signs such as lactate levels, C-reactive protein, and body temperature are widely used for risk assessment. At the same time, the integration of real-time monitoring systems and electronic health records is also constantly advancing to achieve more accurate early warning and intervention.

[0003] When predicting and analyzing the risk of sepsis based on machine learning models, since the symptoms of sepsis have similar signs and symptoms to some other diseases, if the test data involved in the analysis do not have sufficient characterization characteristics during early risk assessment, problems such as distribution shift, data missing, or data drift may occur, affecting the accuracy of the analysis of the prediction model and causing the prediction assessment to still have certain errors. Summary of the invention

[0004] The purpose of the present invention is to provide a system and method for predicting the early risk of sepsis. The technical solutions adopted are as follows:

[0005] The present invention provides a method for predicting the early risk of sepsis, the method comprising:

[0006] Obtain clinical manifestations and values ​​of several types of physical signs for different types of patients; different types of patients include: patients with sepsis and patients without sepsis;

[0007] According to the differences in clinical manifestations of different types of patients, combined with the correlation between each sign detection type and clinical manifestations of sepsis patients, the performance correlation of each sign detection type was obtained;

[0008] According to the difference in the numerical distribution of each physical sign detection type in different types of patients, the stability of the numerical distribution of each physical sign detection type in sepsis patients and the performance correlation of each physical sign detection type, the importance of each physical sign detection type in sepsis patients was obtained;

[0009] A sepsis risk model was obtained based on the importance and value distribution of each sign detection type of the patient.

[0010] Furthermore, the method for obtaining the performance correlation degree includes:

[0011] All clinical manifestations were clustered to obtain representative clusters;

[0012] For any physical sign detection type of sepsis patients, according to the correlation stability of the values ​​of the physical sign detection type of all sepsis patients in each characterization cluster and the corresponding clinical manifestations, the correlation index of the physical sign detection type in each characterization cluster is obtained;

[0013] The overall difference in clinical manifestations between patients with sepsis and patients without sepsis in each representative cluster is used as the performance indicator of each representative cluster;

[0014] According to the association index of the physical sign detection type in each characterization cluster, combined with the performance ability index of each characterization cluster, the performance association degree of the physical sign detection type is obtained.

[0015] Furthermore, the method for obtaining the correlation index includes:

[0016] In any one of the representation clusters, a sum vector of the value of each sepsis patient in the physical sign detection type and the corresponding clinical manifestation is obtained as a joint vector of each sepsis patient in the representation cluster corresponding to the physical sign detection type;

[0017] Calculate the Euclidean distance between each joint vector and each other joint vector in the characterization cluster, and calculate the variance of all Euclidean distances to obtain the distribution disorder of each joint vector; negatively correlate the mean of the distribution disorder of all joint vectors in the characterization cluster to obtain the joint stability index of the physical sign detection type in the characterization cluster;

[0018] Arranging the numerical values ​​of the physical sign detection type of all the sepsis patients in the characterization cluster to obtain a numerical sequence of the characterization cluster; arranging the clinical manifestations of all the sepsis patients in the characterization cluster according to the order of the corresponding numerical sequence to obtain a manifestation sequence of the characterization cluster;

[0019] Calculate the correlation between the numerical sequence and the expression sequence of the characterization cluster, and obtain the relevant index of the physical sign detection type in the characterization cluster;

[0020] The associated index of the physical sign detection type in the characterization cluster is obtained by combining the joint stability index and the related index of the physical sign detection type in the characterization cluster.

[0021] Furthermore, the method for obtaining the performance ability index includes:

[0022] For any characterization cluster, obtain the sum vector of clinical manifestations of all sepsis patients in the characterization cluster as the sepsis manifestation vector of the characterization cluster; obtain the sum vector of clinical manifestations of all non-sepsis patients in the characterization cluster as the non-sepsis manifestation vector of the characterization cluster;

[0023] The Euclidean distance between the sepsis performance vector representing the cluster and the non-sepsis performance vector is used as the performance ability indicator of the cluster representing the cluster.

[0024] Furthermore, the step of obtaining the performance correlation of the physical sign detection type according to the correlation index of each characterization cluster and combining the performance capability index of each characterization cluster to include:

[0025] The product of the association index of the physical sign detection type in each characterization cluster and the performance ability index of the corresponding characterization cluster is used as the adjusted association degree of the physical sign detection type in each characterization cluster;

[0026] The average of the adjusted correlations of the physical sign detection type in all the characterization clusters is taken as the performance correlation of the physical sign detection type.

[0027] Furthermore, the method for obtaining the importance includes:

[0028] For any physical sign detection type of sepsis patients, the performance distinction index of the physical sign detection type is obtained according to the overall numerical difference between sepsis patients and non-sepsis patients on the physical sign detection type;

[0029] According to the consistency of the numerical distribution of the physical sign detection type in the sepsis patients, the performance stability index of the physical sign detection type is obtained;

[0030] The importance of the physical sign detection type is obtained by combining the performance distinction index, performance stability index and performance correlation of the physical sign detection type.

[0031] Furthermore, the method for obtaining the performance differentiation index includes:

[0032] Calculate the mean of the values ​​of all patients with sepsis in this sign detection type to obtain the mean of the diseased patients in this sign detection type; calculate the mean of the values ​​of all patients without sepsis in this sign detection type to obtain the mean of the diseased patients in this sign detection type;

[0033] The difference between the mean value of the diseased group and the mean value of the non-diseased group of the physical sign detection type is used as the performance differentiation index of the physical sign detection type.

[0034] Furthermore, the method for obtaining the performance stability index includes:

[0035] The variance of the values ​​of this physical sign detection type for all sepsis patients is negatively correlated and mapped to obtain a performance stability index for this physical sign detection type.

[0036] Furthermore, the method for obtaining the sepsis risk model includes:

[0037] When the importance of a physical sign detection type is greater than a preset importance threshold, the corresponding physical sign detection type is used as an evaluation type;

[0038] The importance of the patient's assessment type is used as the weight, and the numerical value of the assessment type is used as the training set to input and output into the logistic regression model for training, so as to obtain a trained sepsis risk model; the input of the sepsis risk model is the numerical value of the patient's physical sign detection type, and the output is the risk assessment result.

[0039] The present invention also provides a system for predicting the early risk of sepsis, comprising a memory and a processor, wherein the processor executes a computing program stored in the memory to implement the method for predicting the early risk of sepsis as described above.

[0040] The present invention has the following beneficial effects:

[0041] The present invention analyzes a variety of physical sign detection types of patients with sepsis and patients without sepsis, comprehensively analyzes the importance of each physical sign detection type of sepsis patients based on the correlation between the physical sign detection type and the clinical manifestation, combined with the differentiation strength of sepsis patients and patients without sepsis in different physical sign detection types, and the numerical distribution of sepsis patients in the physical sign detection types, and jointly analyzes the importance of the impact of different physical sign detection types of sepsis patients on the detection through the characterization of clinical manifestations by the physical sign detection type, the specific differentiation strength of the physical sign detection type, and the characterization stability of the physical sign detection type, and obtains the importance of each physical sign detection type. The sepsis risk model obtained by adjusting the importance influences makes the risk prediction accuracy more accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A flow chart of a method for predicting early risk of sepsis provided by one embodiment of the present invention;

[0044] Figure 2 A flow chart of a method for obtaining a degree of correlation provided by an embodiment of the present invention;

[0045] Figure 3 A flow chart of a method for obtaining importance provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the prediction system and method of the early risk of sepsis proposed by the present invention, its specific implementation method, structure, characteristics and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0047] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0048] The specific scheme of a system and method for predicting the early risk of sepsis provided by the present invention is described in detail below with reference to the accompanying drawings.

[0049] See also Figure 1 , which shows a flow chart of a method for predicting early risk of sepsis provided by an embodiment of the present invention, the method comprising the following steps:

[0050] S1: Obtain the clinical manifestations of different types of patients and the values ​​of several types of physical sign tests; different types of patients include: patients with sepsis and patients without sepsis.

[0051] Since sepsis is difficult to diagnose clinically in its early stages, and it develops quickly and is similar to other diseases, early assessment, prediction, and early warning are required. In an embodiment of the present invention, patient information is collected from various medical and health service centers using big data technology. The patients mainly include sepsis patients and non-sepsis patients. The collected information includes the test results and clinical manifestations of the patients under various physical sign detection types, which facilitates subsequent analysis and evaluation based on the different specific characterization levels of the physical sign detection types.

[0052] The physical sign detection types include: blood test: white blood cell count, red blood cell count, platelet count, blood lactate and absolute neutrophil value, etc.; blood gas analysis: pH value, carbon dioxide partial pressure, oxygen partial pressure and bicarbonate, etc.; coagulation function test: inflammatory markers: C-reactive protein, procalcitonin, interleukin-6, amyloid protein and heparin binding protein, etc.; liver function test: alanine aminotransferase, aspartate aminotransferase, total bilirubin and direct bilirubin, etc.; renal function test: urea nitrogen and creatinine, etc.; physiological monitoring: blood pressure, heart rate, respiratory rate, body temperature and blood oxygen saturation, etc. The numerical data of various feature detection types reflect the different detection sign states of patients. When the risk of sepsis is directly analyzed and evaluated through the data of physical sign detection types, some types that are not clearly characterized or have low specificity in physical sign detection types with other diseases will affect the accuracy of risk assessment prediction. Therefore, it is necessary to further analyze the characterization ability of physical sign detection types.

[0053] Among them, clinical manifestations refer to a series of abnormal changes in the patient's body after contracting a certain disease, such as cough, fever, headache, abdominal pain, weakness or hemoptysis, etc. The clinical manifestations reflect the manifestation of the patient's physical symptoms. For patients with sepsis, the more significant the data changes of the sign detection type are reflected in the clinical manifestations, the more critical the analysis of the sign detection type is. When the data changes of the sign detection type make the clinical manifestations of sepsis patients change more drastically, it means that the data changes are more significant in clinical manifestations. For example, when the sign detection type is blood lactate level, the increase in blood lactate level is often related to tissue hypoxia and metabolic disorders. The normal blood lactate level is 0.5-1.7mmol / L. With the significant increase in blood lactate level, sepsis patients show symptoms such as hypotension and confusion. When the blood lactate level reaches 3mmol / L, sepsis patients show septic shock, which reflects that the data changes of blood lactate level have a high correlation with the severity of clinical manifestations, that is, the data changes of blood lactate level are reflected in clinical manifestations. Very significant. When the physical sign detection type is C-reactive protein, C-reactive protein is an acute phase reactive protein used to assess the level of infection. Normal C-reactive protein is within 10 mg / L. As the C-reactive protein level increases, the systemic inflammatory response of sepsis patients will be more severe. Therefore, the higher the correlation between the C-reactive protein level and the clinical manifestations, the more important the blood lactate level and C-reactive protein level are in the subsequent analysis of sepsis risk prediction.

[0054] In the embodiment of the present invention, considering that big data may be missing when it is collected from the source, the data information of some patients is incomplete, and the clinical manifestations are text content data, which has a lot of irrelevant information and needs further structured processing, the collected big data patient information is cleaned and analyzed. Data cleaning is to clean and remove incomplete data or missing data in the collected data to ensure the accuracy of the collected data. Data cleaning is achieved by missing value processing. It should be noted that when obtaining the clinical manifestations of different types of patients and the values ​​of several types of physical sign detection types, the number of data items obtained for each patient is the same, such as each patient needs to obtain the same name, clinical manifestations, white blood cell count, red blood cell count, platelet count and neutrophil absolute value and other information. When any item of patient information is not filled in, such as when the white blood cell count is not filled in, the patient's information has missing items. The information of patients with missing items is cleared because when there are missing items, the credibility of other reference data of the patient is reduced. More specifically: obtain the information sequence corresponding to each patient. The elements in the information sequence are the patient's clinical manifestations and the numerical values ​​of several types of physical sign detection. When there are missing items in the information sequence, that is, there is unfilled information, the information sequence of the patient is deleted. For example, in the information of patients A and B, there is information about clinical manifestations that is not included in the information of patient A, then there are missing items in the information sequence corresponding to patient A. Delete the entire information sequence of this patient to achieve data cleaning. For the text data of clinical manifestations, data analysis is performed after data cleaning. Data analysis is to parse the original, unstructured data into structured data that is easy to process, so that big data can be processed more efficiently and conveniently.

[0055] Therefore, in the embodiment of the present invention, the clinical manifestations are parsed and processed into vectors to achieve data parsing. The NER model is used to identify key entities in the text, such as symptom names, body parts or disease names, etc. On the basis of entity recognition, the features of the entity are further extracted, including the context information of the entity, modifiers, etc. The relationship between the entities in the text is identified by training the relationship extraction model, and the word2vec algorithm is used to convert the features of the text representation corresponding to the extracted patient clinical manifestations into vector form.

[0056] For example, the clinical manifestations collected are "The patient complained of frequent chest pain recently, especially after exercise and when emotionally excited, accompanied by palpitations and dyspnea. The electrocardiogram showed irregular heartbeat." (1) First, the NER model may recognize the entities as: Symptom name: chest pain, palpitations, dyspnea; physical activity: after exercise; emotional state: when emotionally excited; disease name: irregular heartbeat. (2) Further entity feature extraction is: chest pain: the context information is "frequent chest pain", and the modifier is "frequent"; after exercise: the context information is "especially after exercise"; when emotionally excited: the context information is "especially when emotionally excited"; palpitations: the context information is "accompanied by palpitations", and the modifier is "accompanied"; dyspnea: the context information is "accompanied by dyspnea", and the modifier is "accompanied"; irregular heartbeat: the context information is "electrocardiogram shows irregular heartbeat", and the modifier is "electrocardiogram shows". (3) Further identify the relationship between text and entities: chest pain has a causal relationship (trigger condition) with post-exercise and emotional excitement; chest pain has a concurrent relationship with palpitations and dyspnea; electrocardiogram has a detection relationship with arrhythmia. (4) Each entity and its contextual information or modifiers are used as input, and a vector representation is obtained through the word2vec model, such as the vector of chest pain [0.1,-0.2,0.3,...], the vector of palpitations [-0.3,0.2,0.1,...], the vector of dyspnea [0.1,-0.4,0.2,...], and so on.

[0057] It should be noted that the NER model is a natural language processing model that can identify entities with specific meanings from text. The word2vec algorithm is a method for converting words into vector form. The method of parsing text data into vectors is a well-known technical means familiar to those skilled in the art and will not be elaborated here.

[0058] It should be noted that in the specific implementation of this application, the collection, use and processing of patient condition-related data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0059] S2: According to the differences in clinical manifestations among different types of patients, combined with the correlation between each sign detection type and clinical manifestations of sepsis patients, the performance correlation of each sign detection type is obtained.

[0060] The stronger the ability of the clinical manifestations of the disease to express sepsis and the higher the correlation with the data of the sign detection type, the more reliable and necessary it is to analyze the risk of sepsis through this sign detection type. Therefore, we first analyze the correlation of each sign detection type based on the patient's clinical manifestations.

[0061] Preferably, in some possible implementations of the present invention, the method for obtaining the performance correlation degree can be found in Figure 2 , which shows a flow chart of a method for obtaining a performance correlation provided by an embodiment of the present invention, the method comprising the following steps:

[0062] S201: Cluster all clinical manifestations to obtain characteristic clustering clusters.

[0063] Different disease conditions may also have similar symptoms in clinical manifestations. Therefore, analyzing the relevant conditions of the same clinical manifestation through the numerical changes of the physical sign detection type can reflect the degree of correlation between the detection data and the symptoms. The more influential physical sign detection types can increase the importance of analysis. In addition, considering that the clinical manifestations may have different recording habits, resulting in inconsistencies in the extraction of the same symptoms, the clinical manifestations are first divided into similar categories through clustering.

[0064] In an embodiment of the present invention, a K-means clustering algorithm is used to perform clustering according to the similarity of clinical manifestations to obtain characterizing cluster clusters, and the similarity between clinical manifestations is obtained by calculating the similarity between text vectors. It should be noted that clustering and similarity calculation are technical means well known to those skilled in the art, such as using cosine similarity to perform similarity analysis between vectors, etc., which will not be elaborated here.

[0065] S202: For any physical sign detection type of sepsis patients, obtain a correlation index of the physical sign detection type in each characterization cluster according to the correlation stability of the values ​​of the physical sign detection type of all sepsis patients in each characterization cluster and the corresponding clinical manifestations.

[0066] In similar clinical manifestations, when the physical sign detection type data of sepsis patients is more correlated and stable with the clinical manifestations, it means that the symptoms of sepsis have a more relevant influence on this type of physical sign detection type, and higher attention should be paid to this type of physical sign detection type.

[0067] Preferably, in an embodiment of the present invention, the method for obtaining the correlation index includes:

[0068] First, in any characterization cluster, the sum vector of the value of each sepsis patient in the sign detection type and the corresponding clinical manifestation is obtained as the joint vector of each sepsis patient in the characterization cluster corresponding to the sign detection type. The value of the sign detection type can be analyzed as a single-dimensional vector, and the value and clinical manifestation of the sepsis patient under the sign detection type are combined to reflect the joint distribution of the sign detection type through the sum vector.

[0069] Furthermore, the Euclidean distance between each joint vector and each other joint vector in the characterization cluster is calculated, and the variance of all Euclidean distances is calculated to obtain the distribution disorder of each joint vector, reflecting the distribution deviation between a single joint distribution and other joint vectors. The mean of the distribution disorder of all joint vectors in the characterization cluster is negatively correlated to obtain the joint stability index of the sign detection type in the characterization cluster, and the deviation degree of all joint distributions is comprehensively considered. The lower the deviation degree, the more stable the joint distribution is, and the higher the reliability in correlation analysis.

[0070] It should be noted that the Euclidean distance and negative correlation mapping are technical means well known to those skilled in the art. The negative correlation mapping may be in the form of inverse proportion or negative exponential power, etc., which will not be elaborated here.

[0071] Furthermore, the numerical values ​​of all sepsis patients in the characterization cluster in the physical sign detection type are arranged to obtain the numerical sequence of the characterization cluster, and the clinical manifestations of all sepsis patients in the characterization cluster are arranged according to the order of the corresponding numerical sequence to obtain the performance sequence of the characterization cluster. Arranging the data facilitates subsequent analysis of related situations.

[0072] By calculating the correlation between the numerical sequence and the performance sequence of the characterizing cluster cluster, the relevant index of the physical sign detection type in the characterizing cluster cluster is obtained. In the embodiment of the present invention, the relevant index is obtained by calculating the Spearman rank correlation coefficient of the numerical sequence and the performance sequence. As an example, the expression for obtaining the relevant index is:

[0073] ; In the formula, Expressed as Type of physical sign detection The relevant indicators that characterize the clusters are The number of elements represented as a numeric sequence, The first elements, Represented as a numerical sequence The element and the expression sequence The difference of elements.

[0074] When the correlation index is larger, it means that there is a more likely correlation between the physical sign detection type and the clinical manifestation, and the data change of the physical sign detection type is closer to the change of the clinical manifestation. Among them, the vector data dimension in the performance sequence is relatively high. In order to improve the calculation efficiency, the vector in the performance sequence is reduced in dimension. In the embodiment of the present invention, the dimension reduction can be performed by the principal component analysis method. The dimension reduction method is not described here.

[0075] It should be noted that the Spearman rank correlation coefficient is used to calculate the correlation between sequences. The calculation of sequence correlation is a technical means well known to those skilled in the art. For example, the Pearson correlation coefficient can also be used for calculation, which will not be elaborated here.

[0076] Finally, the associated index of the physical sign detection type in the characterization cluster is obtained by combining the joint stability index and the related index of the physical sign detection type in the characterization cluster. In an embodiment of the present invention, the product of the joint stability index and the related index of the physical sign detection type in the characterization cluster is used as the associated index of the physical sign detection type in the characterization cluster. The larger the joint stability index and the related index, the more relevant and stable the physical sign detection type data is to the clinical manifestation.

[0077] S203: The overall difference in clinical manifestations between patients with sepsis and patients without sepsis in each representation cluster is used as a performance capability indicator for each representation cluster.

[0078] The more obvious the overall deviation between patients with sepsis and those without sepsis in the cluster representation, the more expressive the clinical manifestation characteristics in the cluster representation are, and the higher the weight of the cluster representation association analysis should be.

[0079] Preferably, in an embodiment of the present invention, the method for obtaining the performance ability index includes:

[0080] First, for any characterization cluster, the sum vector of the clinical manifestations of all sepsis patients in the characterization cluster is obtained as the sepsis manifestation vector of the characterization cluster. The sum vector of the clinical manifestations of all non-sepsis patients in the characterization cluster is obtained as the non-sepsis manifestation vector of the characterization cluster. All clinical manifestations corresponding to different types of patients are comprehensively analyzed.

[0081] The Euclidean distance between the sepsis performance vector and the non-sepsis performance vector of the cluster characterizing the cluster is further used as the performance ability index of the cluster characterizing the cluster. The larger the performance ability index is, the greater the overall difference between different types of patients is, and the clinical manifestations of different types are more significant.

[0082] S204: Obtaining the performance correlation of the physical sign detection type according to the association index of each characterization cluster and the performance capability index of each characterization cluster.

[0083] The correlation between comprehensive sign detection types and different clinical manifestations, that is, the correlation between comprehensive sign detection types in all representation clusters, reflects the degree to which sign detection types are affected by multiple clinical manifestations. The higher the correlation, the higher the proportion of the corresponding sign detection type in subsequent analysis.

[0084] In an embodiment of the present invention, the product of the association index of the sign detection type in each characterization cluster and the performance ability index of the corresponding characterization cluster is used as the adjusted association degree of the sign detection type in each characterization cluster, combined with the performance ability and association index of the sign detection type in each characterization cluster, to reflect the association influence of the sign detection type with sepsis tendency.

[0085] Then, the mean of the adjusted correlation of the sign detection type in all the characterization clusters is used as the performance correlation of the sign detection type, and combined with the analysis of the sign detection type for all clinical manifestations, a more reliable performance correlation is obtained. As an example, the expression of the performance correlation is:

[0086] ; Expressed as Correlation of performance between the types of physical signs detected, It is expressed as the total number of clusters. Expressed as A performance index that characterizes the clustering ability. Expressed as Type of physical sign detection The correlation index that characterizes the clusters, Expressed as Type of physical sign detection The adjusted associations that characterize the clusters.

[0087] S3: Based on the differences in the numerical distribution of different types of patients on each type of physical sign detection, the stability of the numerical distribution of each type of physical sign detection in sepsis patients, and the performance correlation of each type of physical sign detection, the importance of each type of physical sign detection in sepsis patients is obtained.

[0088] When predicting risks, training and analysis are often performed based on abnormal conditions in the test data. Different types of vital sign detection have different degrees of importance. Under the multi-dimensional information analysis of big data, it is even more necessary to analyze the vital sign detection types that have an important impact in order to improve the accuracy of the prediction.

[0089] Therefore, the importance of each sign detection type is analyzed. The more significant the difference in the value of the sign detection type between patients with sepsis and those without sepsis, the more important the corresponding sign detection type is for risk analysis. When the value of the sign detection type is more stably distributed within the range of patients with sepsis, it means that the sign detection type is more reliable for analysis. Finally, the higher the correlation degree representing the clinical manifestations, the more it needs to be involved in the risk analysis of sepsis.

[0090] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the importance degree can be found in Figure 3 , which shows a flow chart of a method for obtaining importance provided by an embodiment of the present invention, the method comprising the following steps:

[0091] S301: For any physical sign detection type of sepsis patients, according to the overall numerical difference between sepsis patients and non-sepsis patients in the physical sign detection type, obtain a performance differentiation index of the physical sign detection type.

[0092] First, the degree of difference in the overall numerical value reflects the degree of difference between the numerical value of this physical sign detection type in patients with sepsis and those without sepsis. The more obvious the difference is, the better the effect of sepsis risk prediction through this physical sign detection type is, and the higher the differentiation power is.

[0093] Preferably, in an embodiment of the present invention, the method for obtaining the performance differentiation index includes:

[0094] The difference in the distribution of the mean was used to distinguish, and the mean of the values ​​of all patients with sepsis in this sign detection type was calculated to obtain the diseased mean of this sign detection type. The mean of the values ​​of all patients without sepsis in this sign detection type was calculated to obtain the non-disease mean of this sign detection type.

[0095] The difference between the diseased mean and the non-disease mean of the physical sign detection type is used as the performance differentiation index of the physical sign detection type. The overall distribution of the values ​​is reflected by the mean. The greater the overall distribution difference, the greater the performance differentiation index, indicating a higher differentiation strength. In other embodiments of the present invention, the numerical mode of the physical sign detection types of different types of patients can also be used for difference analysis to reflect the difference in the overall numerical distribution, which will not be elaborated here.

[0096] S302: Obtain a performance stability index of the physical sign detection type according to the degree of consistency of the numerical distribution of the sepsis patients in the physical sign detection type.

[0097] When the numerical fluctuation of the same test type is small, it means that different sepsis patients are more stable in this sign test type, and the effect of subsequent analysis and prediction is more stable and better. Therefore, the more stable the numerical distribution of the same sign test type, the higher the demand for this sign test type to participate in subsequent analysis.

[0098] In an embodiment of the present invention, the variance of the values ​​of the physical sign detection type of all sepsis patients is negatively correlated to obtain a performance stability index of the physical sign detection type. The smaller the variance, the smaller the fluctuation and the higher the stability. Therefore, the larger the performance stability index, the more important the corresponding physical sign detection type.

[0099] S303: Obtain the importance of the physical sign detection type by combining the performance distinction index, the performance stability index and the performance correlation of the physical sign detection type.

[0100] Finally, the three aspects are combined to comprehensively reflect the importance of the physical sign detection type in the subsequent analysis. In the embodiment of the present invention, the product of the performance distinction index, the performance stability index and the performance correlation of the physical sign detection type is normalized to obtain the importance of the physical sign detection type. The higher the importance of the physical sign detection type, the higher the attention in the subsequent risk analysis. As an example, the expression of the importance of the physical sign detection type is:

[0101] ; In the formula, Expressed as The importance of the type of physical sign detection, Expressed as Correlation of performance between the types of physical signs detected, Expressed as The performance differentiation indicators of the types of physical sign detection are Expressed as The performance stability indicators of the various types of physical sign detection are Expressed as a normalized processing function.

[0102] It should be noted that normalization is a technical means well known to those skilled in the art, and the normalization method may be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0103] S4: Obtain a sepsis risk model based on the importance and value distribution of each type of vital sign detection for the patient.

[0104] By collecting data on the patient's physical sign detection types to build a risk model, it is convenient for early risk prediction based on the patient's physical sign detection results. The importance of the physical sign detection type is used to weight the attention of abnormal analysis, so that more specific indicator data receives more attention and the accuracy of the analysis is improved.

[0105] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the sepsis risk model includes:

[0106] When the importance of a sign detection type is greater than a preset importance threshold, the corresponding sign detection type will be used as an evaluation type. First, the sign detection types participating in the model construction will be screened, and the sign detection types with higher importance will be used as the evaluation types for subsequent evaluation and analysis.

[0107] Further, the importance of the patient's assessment type is used as a weight, and the value of the assessment type is used as a training set to be input and output into the logistic regression model for training, so as to obtain a trained sepsis risk model, and the input of the sepsis risk model is the value of the patient's physical sign detection type, and the risk assessment result is output. In the embodiment of the present invention, the cross entropy loss function is selected for training, and the gradient descent method is used as the optimization algorithm for training the logistic regression model.

[0108] It should be noted that in the logistic regression model, the loss function is adjusted through different importance weights, and different loss items are weighted, so that the model will pay more attention to those evaluation types with higher weights during the optimization process.

[0109] In other embodiments of the present invention, it can also be input into a tree-based model, such as a random forest or gradient boosting tree, for training, and the model training is performed by using different importance weights of the evaluation type as prior knowledge, that is, the evaluation type features with higher weights are first split. It should be noted that the model training process is a technical means well known to those skilled in the art, and the training data is only weighted by importance here, so the training process is not limited or elaborated here.

[0110] In an embodiment of the present invention, the patient inputs the data of the detected physical sign detection type into the sepsis risk model, and the sepsis risk model outputs a risk assessment result. Subsequently, the patient's physical sign detection type data and the sepsis risk assessment result can be uploaded to the terminal together to provide data support for early risk prediction.

[0111] It should be noted that, in order to facilitate calculation, all indicator data involved in the calculation in the embodiment of the present invention are preprocessed to eliminate the dimension effect. The specific means of eliminating the dimension effect are technical means well known to those skilled in the art and are not limited here.

[0112] In summary, the present invention analyzes a variety of physical sign detection types for patients with sepsis and patients without sepsis, and comprehensively analyzes the importance of each physical sign detection type for patients with sepsis based on the correlation between the physical sign detection type and the clinical manifestations, combined with the differentiation strength of different physical sign detection types between patients with sepsis and patients without sepsis, and the numerical distribution of physical sign detection types for patients with sepsis. The importance of the impact of different physical sign detection types on the detection of patients with sepsis is jointly analyzed through the characterization of clinical manifestations by physical sign detection types, the specific differentiation strength of physical sign detection types, and the characterization stability of physical sign detection types, and the importance of each physical sign detection type is obtained. The sepsis risk model obtained by adjusting the importance influences makes the risk prediction accuracy more accurate and reliable.

[0113] The present invention also provides a system for predicting the early risk of sepsis, comprising a memory and a processor, wherein the processor executes a computing program stored in the memory to implement the method for predicting the early risk of sepsis as described above.

[0114] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying 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 also possible or may be advantageous.

[0115] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for predicting the early risk of sepsis, characterized in that: The method comprises: Obtain clinical manifestations and values ​​of several types of physical signs for different types of patients; different types of patients include: patients with sepsis and patients without sepsis; According to the differences in clinical manifestations of different types of patients, combined with the correlation between each sign detection type and clinical manifestations of sepsis patients, the performance correlation of each sign detection type was obtained; According to the difference in the numerical distribution of each physical sign detection type in different types of patients, the stability of the numerical distribution of each physical sign detection type in sepsis patients and the performance correlation of each physical sign detection type, the importance of each physical sign detection type in sepsis patients was obtained; Obtain a sepsis risk model based on the importance and value distribution of each type of vital sign detection in the patient; The method for obtaining the performance correlation degree includes: All clinical manifestations were clustered to obtain representative clusters; For any physical sign detection type of sepsis patients, according to the correlation stability of the values ​​of the physical sign detection type of all sepsis patients in each characterization cluster and the corresponding clinical manifestations, the correlation index of the physical sign detection type in each characterization cluster is obtained; The overall difference in clinical manifestations between patients with sepsis and patients without sepsis in each representative cluster is used as the performance indicator of each representative cluster; According to the association index of the physical sign detection type in each representation cluster, combined with the performance ability index of each representation cluster, the performance association degree of the physical sign detection type is obtained; The method for obtaining the associated index includes: In any one of the representation clusters, a sum vector of the value of each sepsis patient in the physical sign detection type and the corresponding clinical manifestation is obtained as a joint vector of each sepsis patient in the representation cluster corresponding to the physical sign detection type; Calculate the Euclidean distance between each joint vector and each other joint vector in the characterization cluster, and calculate the variance of all Euclidean distances to obtain the distribution disorder of each joint vector; negatively correlate the mean of the distribution disorder of all joint vectors in the characterization cluster to obtain the joint stability index of the physical sign detection type in the characterization cluster; Arranging the numerical values ​​of the physical sign detection type of all the sepsis patients in the characterization cluster to obtain a numerical sequence of the characterization cluster; arranging the clinical manifestations of all the sepsis patients in the characterization cluster according to the order of the corresponding numerical sequence to obtain a manifestation sequence of the characterization cluster; Calculate the correlation between the numerical sequence and the expression sequence of the characterization cluster, and obtain the relevant index of the physical sign detection type in the characterization cluster; Combining the joint stability index and the related index of the physical sign detection type in the characterization cluster, obtaining the associated index of the physical sign detection type in the characterization cluster; The method for obtaining the performance ability indicator includes: For any characterization cluster, obtain the sum vector of clinical manifestations of all sepsis patients in the characterization cluster as the sepsis manifestation vector of the characterization cluster; obtain the sum vector of clinical manifestations of all non-sepsis patients in the characterization cluster as the non-sepsis manifestation vector of the characterization cluster; The Euclidean distance between the sepsis performance vector representing the cluster and the non-sepsis performance vector is used as the performance ability indicator of the cluster; The step of obtaining the performance correlation of the physical sign detection type according to the correlation index of each characterization cluster and the performance capability index of each characterization cluster includes: The product of the association index of the physical sign detection type in each characterization cluster and the performance ability index of the corresponding characterization cluster is used as the adjusted association degree of the physical sign detection type in each characterization cluster; The average of the adjusted correlations of the physical sign detection type in all the representation clusters is used as the performance correlation of the physical sign detection type; The method for obtaining the importance includes: For any physical sign detection type of sepsis patients, the performance distinction index of the physical sign detection type is obtained according to the overall numerical difference between sepsis patients and non-sepsis patients on the physical sign detection type; According to the consistency of the numerical distribution of the physical sign detection type in the sepsis patients, the performance stability index of the physical sign detection type is obtained; The importance of the physical sign detection type is obtained by combining the performance distinction index, the performance stability index and the performance correlation of the physical sign detection type; The method for obtaining the performance differentiation index includes: Calculate the mean of the values ​​of all patients with sepsis in this sign detection type to obtain the mean of the diseased patients in this sign detection type; calculate the mean of the values ​​of all patients without sepsis in this sign detection type to obtain the mean of the diseased patients in this sign detection type; The difference between the mean of the diseased and the mean of the non-disease-affected patients of the physical sign detection type is used as the performance differentiation index of the physical sign detection type; The method for obtaining the performance stability index includes: The variance of the values ​​of this physical sign detection type for all sepsis patients is negatively correlated and mapped to obtain a performance stability index for this physical sign detection type.

2. A method for predicting early risk of sepsis according to claim 1, characterized in that: The method for obtaining the sepsis risk model includes: When the importance of a physical sign detection type is greater than a preset importance threshold, the corresponding physical sign detection type is used as an evaluation type; The importance of the patient's assessment type is used as the weight, and the numerical value of the assessment type is used as the training set to input and output into the logistic regression model for training, so as to obtain a trained sepsis risk model; the input of the sepsis risk model is the numerical value of the patient's physical sign detection type, and the output is the risk assessment result.

3. A system for predicting the early risk of sepsis, comprising a memory and a processor, characterized in that: The processor executes the computing program stored in the memory to implement the method for predicting the early risk of sepsis as claimed in any one of claims 1 to 2.

Images