CN118899088B - A system and method for predicting early risk of sepsis - Google Patents
A system and method for predicting early risk of sepsis Download PDFInfo
- Publication number
- CN118899088B CN118899088B CN202411398140.9A CN202411398140A CN118899088B CN 118899088 B CN118899088 B CN 118899088B CN 202411398140 A CN202411398140 A CN 202411398140A CN 118899088 B CN118899088 B CN 118899088B
- Authority
- CN
- China
- Prior art keywords
- sign detection
- sepsis
- detection type
- index
- characterization
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Pathology (AREA)
- Artificial Intelligence (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention relates to the technical field of health data prediction, in particular to a sepsis early risk prediction system and method. The method comprises the steps of obtaining the performance association degree of each physical sign detection type through the clinical performance difference condition of a patient suffering from sepsis and a patient not suffering from sepsis and the association condition between a plurality of physical sign detection types and clinical performance, and obtaining the importance of each physical sign detection type by combining the distinguishing strength of the patient suffering from sepsis and the patient not suffering from sepsis in different physical sign detection types and the numerical distribution of the patient suffering from sepsis in the physical sign detection types, and obtaining a sepsis risk model based on the importance of the physical sign detection types and the numerical value. According to the invention, the importance degree is reflected by the specific distinguishing degree and the characterization stability degree of the patient sign detection type and the characterization condition of clinical manifestation, and the risk prediction accuracy is more accurate and reliable by performing more accurate model acquisition through the importance degree.
Description
Technical Field
The invention relates to the technical field of health data prediction, in particular to a sepsis early risk prediction system and method.
Background
Sepsis is a general term for surgical infections with systemic inflammatory response, such as changes in body temperature, respiration, circulation, etc. The current situation of early risk prediction of sepsis mainly depends on clinical monitoring data and analysis of biomarkers after endotoxin and exotoxin generated by pathogenic bacteria and various inflammatory mediators mediated by the endotoxin and the exotoxin are absorbed to body tissues. For example, indicators of lactate levels, C-reactive protein, and body temperature are widely used for risk assessment. Meanwhile, integration of a real-time monitoring system and an electronic health record is also continuously advanced, so that more accurate early warning and intervention are realized.
When the machine learning model is used for predicting and analyzing sepsis risks, as symptoms of sepsis have similar symptoms with some other diseases, problems such as distribution deviation, data missing or data drifting can occur if the characterization characteristics of detection data involved in analysis are insufficient in early risk assessment, so that the accuracy of analysis of the prediction model is affected, and the prediction assessment still has a certain error.
Disclosure of Invention
The invention aims to provide a sepsis early risk prediction system and a sepsis early risk prediction method, and the adopted technical scheme is as follows:
The invention provides a method for predicting early risk of sepsis, which comprises the following steps:
Obtaining clinical manifestations of different types of patients and numerical values of a plurality of sign detection types, wherein the different types of patients comprise sepsis patients and patients without sepsis;
according to the difference degree of different types of patients on clinical manifestations, combining the correlation degree between each sign detection type of sepsis patients and the clinical manifestations to obtain the performance correlation degree of each sign detection type;
Obtaining the importance of each physical sign detection type of the sepsis patient according to the numerical distribution difference condition of different types of patients on each physical sign detection type, the numerical distribution stability condition of the sepsis patient on each physical sign detection type and the performance association degree of each physical sign detection type;
Based on the importance and the numerical distribution of the patient in each sign detection type, a sepsis risk model is obtained.
Further, the method for obtaining the performance association degree comprises the following steps:
clustering all clinical manifestations to obtain a characterization cluster;
For any sign detection type of sepsis patients, obtaining an associated index of the sign detection type in each characterization cluster according to the numerical value of the sign detection type of all sepsis patients in each characterization cluster and the relevant stability condition of the corresponding clinical manifestation;
according to the overall difference degree of clinical manifestations of patients suffering from sepsis and patients not suffering from sepsis in each characterization cluster, taking the overall difference degree as an expressive power index of each characterization cluster;
And according to the association index of the sign detection type in each characterization cluster, combining the performance capability index of each characterization cluster to obtain the performance association degree of the sign detection type.
Further, the method for acquiring the association index comprises the following steps:
In any characterization cluster, acquiring a sum vector of a numerical value of each sepsis patient in the sign detection type and a corresponding clinical manifestation, and taking the sum vector as a joint vector of each sepsis patient in the characterization cluster, wherein the joint vector corresponds to the sign detection type;
Calculating Euclidean distance between each joint vector and each other joint vector in the characterization cluster, and solving variance of all Euclidean distances to obtain distribution confusion of each joint vector; carrying out negative correlation mapping on the average value of the distribution confusion degree of all the joint vectors in the characterization cluster to obtain a joint stability index of the sign detection type in the characterization cluster;
Arranging the clinical manifestations of all sepsis patients in the characterization cluster according to the sequence of the corresponding numerical sequences to obtain the manifestation sequence of the characterization cluster;
Calculating the correlation between the numerical sequence and the expression sequence of the characterization cluster to obtain the correlation index of the sign detection type in the characterization cluster;
And combining the combined stable index and the related index of the sign detection type in the characterization cluster to obtain the related index of the sign detection type in the characterization cluster.
Further, the method for obtaining the expressive force index comprises the following steps:
For any one characterization cluster, obtaining the sum vector of clinical manifestations of all sepsis patients in the characterization cluster as a sepsis manifestation vector of the characterization cluster; obtaining the sum vector of clinical manifestations of all patients without sepsis in the characterization cluster, and taking the sum vector as the expression vector without sepsis of the characterization cluster;
and taking the Euclidean distance between the sepsis expression vector of the characterization cluster and the sepsis expression vector without sepsis as the expression capacity index of the characterization cluster.
Further, the obtaining the performance association degree of the sign detection type according to the association index of the sign detection type in each characterization cluster and combining the performance capability index of each characterization cluster includes:
Taking the product of the association index of the sign detection type in each characterization cluster and the expressive power index of the corresponding characterization cluster as the adjustment association degree of the sign detection type in each characterization cluster;
and taking the average value of the adjustment association degree of the sign detection type in all the characterization cluster as the performance association degree of the sign detection type.
Further, the method for obtaining the importance degree comprises the following steps:
For any sign detection type of a sepsis patient, obtaining a performance distinguishing index of the sign detection type according to the overall numerical difference degree of the sepsis patient and the non-sepsis patient on the sign detection type;
obtaining a performance stability index of the physical sign detection type according to the consistent degree of the numerical distribution of the sepsis patient on the physical sign detection type;
And combining the performance distinguishing index, the performance stabilizing index and the performance association degree of the sign detection type to obtain the importance degree of the sign detection type.
Further, the method for obtaining the performance differentiation index comprises the following steps:
Calculating the average value of the numerical values of all patients without sepsis in the physical sign detection type to obtain the diseased average value of the physical sign detection type;
and taking the difference between the diseased average value and the non-diseased average value of the sign detection type as a performance distinguishing index of the sign detection type.
Further, the method for obtaining the performance stability index comprises the following steps:
And carrying out negative correlation mapping on the variances of the numerical values of all sepsis patients in the physical sign detection type to obtain the performance stability index of the physical sign detection type.
Further, the method for acquiring the sepsis risk model includes:
When the importance degree of the sign detection type is larger than a preset importance threshold value, the corresponding sign detection type is used as an evaluation type;
Taking the importance of the evaluation type of the patient as a weight, taking the value of the evaluation type as a training set to enter and exit into a logistic regression model for training to obtain a sepsis risk model after training, inputting the sepsis risk model as the value of the sign detection type of the patient, and outputting a risk evaluation result.
The invention also provides a sepsis early risk prediction system, which comprises a memory and a processor, wherein the processor executes a calculation program stored in the memory to realize the sepsis early risk prediction method.
The invention has the following beneficial effects:
According to the invention, through analyzing various physical sign detection types of a patient suffering from sepsis and a patient not suffering from sepsis, the importance degree of each physical sign detection type is obtained by analyzing the importance degree of different physical sign detection types of the patient suffering from sepsis on the basis of the correlation condition between the physical sign detection types and clinical manifestations and combining the distinguishing strength of the patient suffering from sepsis and the patient not suffering from sepsis on different physical sign detection types and the numerical distribution of the patient suffering from sepsis on the physical sign detection types, the importance degree of each physical sign detection type of the patient suffering from sepsis is comprehensively analyzed, the characterization of the physical sign detection types on the clinical manifestation, the specific distinguishing strength of the physical sign detection types and the characterization stability of the physical sign detection types are comprehensively analyzed, the importance degree of the influence of the different physical sign detection types of the patient on the detection is jointly analyzed, and the importance degree of each physical sign detection type is obtained, and the sepsis risk model is obtained by adjusting the importance degree influence.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions and advantages of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for predicting early risk of sepsis according to one embodiment of the present invention;
FIG. 2 is a flowchart of a method for obtaining a performance relevance according to an embodiment of the present invention;
fig. 3 is a flowchart of a method for obtaining importance according to an embodiment of the present invention.
Detailed Description
In order to further describe the technical means and effects of the present invention for achieving the intended purpose, the following detailed description refers to the accompanying drawings and preferred embodiments, in which a sepsis early stage risk prediction system and method according to the present invention is described in detail. In the following description, different "one embodiment" or "another embodiment" means that the embodiments are not necessarily the same. Furthermore, the particular features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.
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.
The following specifically describes a specific scheme of a sepsis early stage risk prediction system and method provided by the present invention with reference to the accompanying drawings.
Referring to fig. 1, a flowchart of a method for predicting early risk of sepsis according to one embodiment of the present invention is shown, the method includes the following steps:
S1, acquiring clinical manifestations of different types of patients and numerical values of a plurality of sign detection types, wherein the different types of patients comprise patients suffering from sepsis and patients not suffering from sepsis.
Because sepsis is relatively difficult in clinical diagnosis in an initial stage, the onset of the sepsis is rapid and similar to other diseases, early evaluation, prediction and early warning are required, in the embodiment of the invention, the information of patients is collected from various medical health service centers through a big data technology, the patients mainly comprise sepsis patients and patients without sepsis, the collected information comprises detection results and clinical manifestations of the patients under various sign detection types, and the follow-up analysis and evaluation are convenient according to different specific characterization degrees of the sign detection types.
The blood gas analysis comprises pH value, partial pressure of carbon dioxide, partial pressure of oxygen, bicarbonate and the like, the blood coagulation function test comprises inflammatory markers such as C-reactive protein, procalcitonin, interleukin-6, amyloid protein, heparin binding protein and the like, the liver function test comprises alanine aminotransferase, aspartic acid aminotransferase, total bilirubin, direct bilirubin and the like, the kidney function test comprises urea nitrogen, creatinine and the like, the physiological monitoring comprises blood pressure, heart rate, respiratory rate, body temperature, blood oxygen saturation and the like, the numerical data of various characteristic detection types reflect different detection sign states of a patient, and when the analysis and evaluation of sepsis risk are directly carried out through the data of the characteristic detection types, the types which are not obvious or have low specificity with other diseases on the characteristic detection types influence the accuracy of risk assessment prediction, so that the further analysis of the characteristic capability of the characteristic detection types is required.
Wherein clinical manifestations refer to a series of abnormal changes in the body of a patient after a certain disease is obtained, such as coughing, fever, headache, abdominal pain, weakness or hemoptysis, etc., the manifestation of a condition of the patient's body is reflected by the clinical manifestations, and the more significant the changes in the data of the sign detection type are reflected on the clinical manifestations, the more critical the sign detection type analysis is. When the data change of the physical sign detection type causes the severe disease change of a sepsis patient in clinical manifestation, the data change is more obvious in clinical manifestation, for example, when the physical sign detection type is blood lactic acid water, the blood lactic acid level is often related to tissue hypoxia and metabolic disorder, the normal blood lactic acid level is 0.5-1.7mmol/L, the sepsis patient presents symptoms such as hypotension and confusion along with the obvious increase of the blood lactic acid level, and when the blood lactic acid level reaches 3mmol/L, the sepsis patient presents septic shock, thereby the data change reflecting the blood lactic acid level has higher related influence on the severity of the clinical manifestation, namely, the data change reflecting the blood lactic acid level is extremely obvious in clinical manifestation. When the physical sign detection type is C-reactive protein, the C-reactive protein is an acute phase reactive protein for evaluating the infection level, the normal C-reactive protein is within 10mg/L, and the systemic inflammatory response of sepsis patients is more serious along with the increase of the C-reactive protein level, so that the higher the correlation effect between the C-reactive protein level and the clinical manifestation is, the more important the blood lactic acid level and the C-reactive protein level are in the follow-up analysis of sepsis risk prediction.
In the embodiment of the invention, considering that the condition of data missing is generated when big data is collected, the data information of partial patients is incomplete and clinically presents as text content data, and more irrelevant information is needed to be further structured, therefore, the collected big data patient information is subjected to data cleaning and data analysis, the data cleaning is to clean and reject the incomplete data or the data of missing items in the collected data so as to ensure the accuracy of the collected data, and the data cleaning is realized through the missing value processing. And when the patient has any item of information unfilled, such as a white blood cell count unfilled, the patient's information has a missing item. The information of the patient having the missing item is cleared because the confidence of the other reference data of the patient is reduced when the missing item is present. More specifically, the information sequence corresponding to each patient is obtained, the elements in the information sequence are the clinical manifestations of the patient and the numerical values of a plurality of sign detection types, and when the missing items exist in the information sequence, namely unfilled information exists in the information sequence, the information sequence of the patient is deleted. For example, if the information of the patient a and the information of the patient b do not contain the information of the clinical characterization, the information sequence corresponding to the patient a has a missing item, and the whole information sequence of the patient is deleted, so that the data cleaning is realized. For the text data of clinical manifestation, after data cleaning, data analysis is to analyze the original unstructured data into structured data which is convenient to process, so that the data can be processed more efficiently and conveniently.
Therefore, in the embodiment of the invention, the clinical manifestation is analyzed and processed into the vector so as to realize data analysis, the clinical manifestation uses the NER model to identify key entities in the text, such as symptom names, body parts or disease names, and the like, on the basis of entity identification, the characteristics of the entities, including the contextual information, modifier words and the like of the entities, are further extracted, the relation among the entities in the text is identified through the training relation extraction model, and the characteristics of the text representation corresponding to the extracted clinical manifestation of the patient are converted into the vector form by using word2vec algorithm.
For example, the clinical manifestations collected are "chest pain is frequently seen recently by patient complaints, especially after exercise and at emotional agitation, with concomitant palpitations and dyspnea. The electrocardiogram shows arrhythmia. "(1) first the NER model may identify entities as symptoms names chest pain, palpitations, dyspnea, physical activity after exercise, emotional states when emotionally activated, and disease names as arrhythmia. (2) The chest pain is further extracted through physical characteristics, namely, chest pain is frequently generated in the context information, the modifier is frequently generated, the context information is particularly after exercise, the context information is particularly at the time of emotion activation, palpitation is accompanied by palpitation, the modifier is accompanied by dyspnea, the context information is accompanied by dyspnea, the modifier is accompanied by heart rate, the arrhythmia is displayed by an electrocardiogram, and the modifier is displayed by the electrocardiogram. (3) Further identifying the relation between the text and the entity, namely causal relation (trigger condition) between chest pain and after exercise and when emotion is excited, concurrence relation between chest pain and palpitation and dyspnea, and detection relation between electrocardiogram and arrhythmia. (4) Vector representations such as chest pain vector [0.1, -0.2,0.3, ], palpitation vector [ -0.3,0.2,0.1, ], dyspnea vector [0.1, -0.4,0.2, ], and the like are obtained by word2vec model with each entity and its context information or modifier as input.
It should be noted that, the NER model is a natural language processing model, and can identify an entity with a specific meaning from a text, the word2vec algorithm is a method for converting a word into a vector form, and a method for analyzing text data into a vector is a well-known means known to those skilled in the art, which is not described herein.
It should be noted that, in the embodiments of the present application, the collection, use and treatment of patient condition related data is required to comply with relevant laws and regulations and standards of relevant countries and regions.
S2, according to the difference degree of different types of patients on clinical manifestations, combining the correlation degree between each sign detection type and the clinical manifestation of the sepsis patients, and obtaining the performance correlation degree of each sign detection type.
When the clinical manifestation of the disease has stronger expression capability on sepsis and higher degree of correlation with the data of the physical sign detection type, the reliability of analyzing the sepsis risk by the physical sign detection type is higher, and the necessity is higher, so that the analysis of the correlation condition of each physical sign detection type is carried out according to the clinical manifestation analysis of the patient.
In some possible implementation manners of the embodiment of the present invention, please refer to fig. 2, which shows a flowchart of a method for obtaining a performance association according to an embodiment of the present invention, where the method includes the following steps:
and S201, clustering all clinical manifestations to obtain a characterization cluster.
Similar symptoms can exist in the on-site manifestations of different symptoms, so that the correlation condition of the same clinical manifestation can be analyzed through the numerical change of the sign detection type, the correlation degree between the detection data and the symptoms can be reflected, and the analysis importance degree can be improved for the sign detection type with influence. And considering that clinical manifestations may exist because of different recording habits, the extraction of the same symptoms is not completely consistent, so the clinical manifestations are firstly classified similarly by clustering.
In the embodiment of the invention, the clustering is performed by adopting a K-means clustering algorithm according to the similarity degree of clinical manifestations to obtain the characterization clustering cluster, and the similarity between the clinical manifestations is obtained through the similarity calculation between text vectors, wherein the clustering and the similarity calculation are technical means well known to those skilled in the art, such as performing the similarity analysis between vectors by adopting cosine similarity, and the like, and are not described herein.
S202, for any sign detection type of sepsis patients, according to the numerical value of the sign detection type of all sepsis patients in each characterization cluster and the relevant stability condition of the corresponding clinical manifestation, obtaining the relevant index of the sign detection type in each characterization cluster.
In a similar clinical presentation, when the sign-detection type data of a patient with sepsis is more relevant and stable to the clinical presentation, it is stated that the symptom-manifestation of sepsis has a more relevant effect on this type of sign-detection type, for which a higher attention should be paid.
Preferably, in an embodiment of the present invention, the method for acquiring the association index includes:
Firstly, in any characterization cluster, a sum vector of a numerical value of each sepsis patient in the sign detection type and a corresponding clinical manifestation is obtained and is used as a joint vector of each sepsis patient in the characterization cluster corresponding to the sign detection type. The numerical value of the sign detection type can be used as a single-dimensional vector for analysis, and the numerical value and clinical manifestation of the patient with the combined sepsis under the sign detection type reflect the combined distribution condition under the sign detection type through the sum vector.
Further, euclidean distance between each joint vector and each other joint vector in the characterization cluster is calculated, variance of all Euclidean distances is calculated, distribution confusion of each joint vector is obtained, and distribution deviation degree between single joint distribution condition and other joint vectors is reflected. And carrying out negative correlation mapping on the average value of the distribution confusion degree of all the joint vectors in the characterization cluster to obtain a joint stability index of the sign detection type in the characterization cluster, and integrating the deviation degree of all the joint distributions, wherein when the lower the deviation degree is, the more stable the joint distribution is, and the higher the reliability degree in the relevance analysis is.
It should be noted that, the euclidean distance and the negative correlation mapping are technical means well known to those skilled in the art, and the negative correlation mapping may be in the form of inverse proportion or negative exponent, and will not be described herein.
Further, the numerical sequence of the characterization cluster is obtained by arranging the numerical values of all sepsis patients in the characterization cluster in the physical sign detection type, and the clinical manifestation of all sepsis patients in the characterization cluster is arranged according to the sequence of the corresponding numerical sequence to obtain the manifestation sequence of the characterization cluster. Subsequent analysis of the relevant conditions is facilitated by ranking the data.
The correlation index of the sign detection type in the characterization cluster is obtained by calculating the correlation of the numerical sequence and the expression sequence of the characterization cluster. As one example, the acquisition expression of the correlation index is:
In the formula (I), in the formula (II), Denoted as the firstThe detection type of the seed sign is in the firstThe associated index that characterizes the cluster,The number of elements represented as a sequence of values,The first of the sequence of values and the sequence of representationsThe number of elements to be added to the composition,Expressed as a sequence of valuesThe first element in the expression sequenceDifferences in individual elements.
When the related index is larger, the more likely the association change relation exists between the sign detection type and the clinical manifestation, and the more compact the data change of the sign detection type and the clinical manifestation change. The vector data in the expression sequence has higher dimension, and the dimension of the vector in the expression sequence is reduced in order to improve the calculation efficiency.
It should be noted that, the spearman level correlation coefficient is used to calculate the correlation between sequences, and the sequence correlation calculation is a technical means well known to those skilled in the art, for example, the spearson correlation coefficient calculation may be also used, which is not described herein.
And finally, combining the combined stable index and the related index of the sign detection type in the characterization cluster to obtain the related index of the sign detection type in the characterization cluster. In the embodiment of the invention, the product of the combined stability index and the related index of the sign detection type in the characterization cluster is used as the related index of the sign detection type in the characterization cluster, and the larger the combined stability index and the related index, the more relevant and stable the sign detection type data and the clinical manifestation situation are.
S203, according to the overall difference degree of clinical manifestations of patients suffering from sepsis and patients not suffering from sepsis in each characterization cluster, taking the overall difference degree as an expressive power index of each characterization cluster.
The more obvious the overall deviation degree of the sepsis patients in the characterization cluster and the sepsis patients not suffering from the sepsis is, the more expressive the clinical manifestation characteristics in the characterization cluster are, and the higher the weight ratio of the analysis association condition of the characterization cluster is.
Preferably, in an embodiment of the present invention, the method for acquiring the performance capability index includes:
Firstly, for any one characterization cluster, obtaining a sum vector of clinical manifestations of all sepsis patients in the characterization cluster, taking the sum vector of clinical manifestations of all non-sepsis patients in the characterization cluster as a sepsis manifestation vector of the characterization cluster, taking the sum vector of clinical manifestations of all non-sepsis patients in the characterization cluster as a non-sepsis manifestation vector of the characterization cluster, and comprehensively carrying out overall analysis on all clinical manifestation conditions corresponding to different types of patients.
Further, the Euclidean distance between sepsis expression vectors of the characterization cluster and the sepsis expression vectors without sepsis is used as the expression capability index of the characterization cluster, and when the expression capability index is larger, the overall difference of patients among different types is larger, and the clinical manifestation among different types is more remarkable.
S204, according to the association indexes of the sign detection type in each characterization cluster, combining the performance capability indexes of each characterization cluster to obtain the performance association degree of the sign detection type.
The correlation condition of the syndrome detection type and different clinical manifestations, namely the correlation condition of the syndrome detection type among all characterization cluster, reflects the expression degree of the syndrome detection type under various clinical manifestations, and when the expression correlation degree is higher, the specific gravity of the corresponding syndrome detection type is higher in the subsequent analysis.
In the embodiment of the invention, the product of the association index of the sign detection type in each characterization cluster and the expression capability index of the corresponding characterization cluster is used as the adjustment association degree of the sign detection type in each characterization cluster, and the association influence of the sign detection type biased by sepsis is reflected by combining the expression capability and the association index of the sign detection type in each characterization cluster.
And then, taking the average value of the adjustment association degree of the sign detection type in all the characterization cluster as the performance association degree of the sign detection type, and combining analysis of the sign detection type on all clinical manifestations to obtain more reliable performance association degree. As one example, the expression that expresses the degree of association is:
; Denoted as the first The degree of performance association of the type of the vital sign detection,Represented as a total number of characterized clusters,Denoted as the firstThe expressive power index of each characterization cluster,Denoted as the firstThe detection type of the seed sign is in the firstThe associated index of each characterization cluster,Denoted as the firstThe detection type of the seed sign is in the firstThe adjustment association degree of each characterization cluster.
And S3, obtaining the importance degree of each physical sign detection type of the sepsis patient according to the numerical distribution difference condition of different types of patients on each physical sign detection type, the numerical distribution stability condition of the sepsis patient on each physical sign detection type and the performance association degree of each physical sign detection type.
Training analysis is often carried out on the basis of abnormal conditions of detection data in risk prediction, different important influence degrees are provided for different sign detection types, and under multi-dimensional information analysis of big data, the sign detection types with important influence are required to be analyzed, so that the accuracy of prediction is improved.
Therefore, the analysis of the importance degree of influence is carried out on each physical sign detection type, when the sepsis patient is more obviously distinguished from the patient without sepsis in terms of the numerical value of the physical sign detection type, the corresponding physical sign detection type is more important for risk analysis, and when the numerical value of the physical sign detection type is distributed more stably in the range of the concentration patients, the condition of the physical sign detection type for analysis is more reliable, and finally, the higher the association degree of the clinical manifestation is combined, the more the patient needs to participate in the risk analysis of sepsis.
In some possible implementation manners of the embodiment of the present invention, referring to fig. 3, a flowchart of a method for obtaining importance according to an embodiment of the present invention is shown, where the method includes the following steps:
s301, for any sign detection type of a sepsis patient, obtaining a performance distinguishing index of the sign detection type according to the overall numerical difference degree of the sepsis patient and the non-sepsis patient on the sign detection type.
Firstly, the degree of difference between the sepsis patients and the non-sepsis patients is reflected by the degree of difference of the whole numerical value, and when the difference is more obvious, the effect of sepsis risk prediction by the physical sign detection type is better and the distinguishing strength is higher.
Preferably, in an embodiment of the present invention, the method for acquiring the performance differentiation index includes:
And distinguishing through the distribution difference of the mean values, calculating the mean value of the numerical values of all patients suffering from sepsis in the physical sign detection type, obtaining the diseased mean value of the physical sign detection type, and calculating the mean value of the numerical values of all patients not suffering from sepsis in the physical sign detection type, thus obtaining the non-diseased mean value of the physical sign detection type.
And taking the difference between the diseased average value and the non-diseased average value of the physical sign detection type as a performance distinguishing index of the physical sign detection type, reflecting the overall distribution of the numerical values through the average value, and indicating that the higher the distinguishing strength is, the larger the overall distribution difference is. In other embodiments of the present invention, the difference analysis may be performed by using the numerical modes of the physical sign detection types of different types of patients, so as to reflect the difference of the overall numerical distribution conditions, which is not described herein.
S302, obtaining the performance stability index of the physical sign detection type according to the consistency degree of the numerical distribution of the sepsis patient on the physical sign detection type.
When the value fluctuation condition of the same detection type is smaller, the condition that different sepsis patients are more stable in the sign detection type is indicated, and the effect for subsequent analysis and prediction is more stable and better, so that the value distribution of the same sign detection type is more stable, and the higher the requirement that the sign detection type participates in subsequent analysis is indicated.
In the embodiment of the invention, the variance of the numerical values of all sepsis patients in the physical sign detection type is mapped in a negative correlation mode, so that the performance stability index of the physical sign detection type is obtained, when the smaller variance indicates the smaller fluctuation degree, the higher stability degree is, and therefore, the higher performance stability index indicates the importance of the corresponding physical sign detection type.
S303, combining the performance distinguishing index, the performance stabilizing index and the performance association degree of the sign detection type to obtain the importance degree of the sign detection type.
Finally, the importance degree of the sign detection type participating in the subsequent analysis is comprehensively reflected by combining three aspects, and in the embodiment of the invention, the product of the performance distinguishing index, the performance stabilizing index and the performance association degree of the sign detection type is normalized to obtain the importance degree of the sign detection type, and the higher the importance degree is, the higher the attention degree is in the subsequent risk analysis. As one example, the expression of the importance of the sign detection type is:
In the formula (I), in the formula (II), Denoted as the firstImportance of the type of the detection of the seed sign,Denoted as the firstThe degree of performance association of the type of the vital sign detection,Denoted as the firstThe performance differentiation index of the type of the detected seed sign,Denoted as the firstThe performance stability index of the type of the detection of the seed sign,Represented as a normalized processing function.
It should be noted that, normalization is a technical means well known to those skilled in the art, and the normalization may be selected from linear normalization, standard normalization, and the like, and the specific normalization method is not limited herein.
And S4, acquiring a sepsis risk model based on the importance degree and the numerical distribution of the patient in each sign detection type.
The risk model is constructed by collecting the sign detection type data of the patient, so that early risk prediction can be conveniently carried out according to the sign detection condition of the patient, the attention degree of the abnormality analysis is weighted by the importance degree of the sign detection type, more attention is obtained by the specific index data, and the analysis accuracy is improved.
Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining a sepsis risk model includes:
When the importance of the sign detection type is larger than a preset importance threshold, the corresponding sign detection type is used as an evaluation type, the sign detection type participating in the model construction is screened at first, and the sign detection type with higher importance is used as an evaluation type for subsequent evaluation analysis.
Further, taking the importance of the evaluation type of the patient as a weight, taking the value of the evaluation type as a training set to enter and exit into a logistic regression model for training, obtaining a sepsis risk model after training, inputting the sepsis risk model as the value of the sign detection type of the patient, and outputting a risk evaluation result. In the embodiment of the invention, a cross entropy loss function is selected for training, and a gradient descent method is adopted as an optimization algorithm for training a logistic regression model.
It should be noted that, in the logistic regression model, the loss function is adjusted by different importance weights, so that the model will pay more attention to those evaluation types with higher weights in the optimization process.
In other embodiments of the present invention, the model training may be performed by inputting the model training to a tree-based model, such as a random forest or a gradient lifting tree, and performing model training by taking different importance weights of the evaluation types as prior knowledge, that is, splitting is performed on the evaluation type features with higher weights. 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 weighted only by importance, so the training process is not limited and described herein.
In the embodiment of the invention, the patient inputs the detected sign detection type data into the sepsis risk model, the sepsis risk model outputs a risk assessment result, and the sign detection type data of the patient and the sealing assessment result can be uploaded to the terminal together later, so that data support for early risk prediction is provided.
It should be noted that, in order to facilitate the operation, all index data involved in the operation in the embodiment of the present invention is subjected to data preprocessing, so as to cancel the dimension effect. The specific means for removing the dimension influence is a technical means well known to those skilled in the art, and is not limited herein.
In summary, the method analyzes the various physical sign detection types of the sepsis patient and the non-sepsis patient, combines the distinguishing strength of the sepsis patient and the non-sepsis patient in different physical sign detection types and the numerical distribution of the sepsis patient in the physical sign detection types from the association condition between the physical sign detection types and the clinical manifestation, comprehensively analyzes the importance of each physical sign detection type of the sepsis patient, analyzes the influence importance degree of the different physical sign detection types of the sepsis patient on the detection by the characterization of the physical sign detection types on the clinical manifestation, the specific distinguishing strength of the physical sign detection types and the characterization stability of the physical sign detection types, and obtains the importance of each physical sign detection type by jointly analyzing the influence importance degree of the different physical sign detection types of the sepsis patient, and enables the risk prediction accuracy to be more accurate and reliable by the sepsis risk model obtained by the influence adjustment of the importance degree.
The invention also provides a sepsis early risk prediction system, which comprises a memory and a processor, wherein the processor executes a calculation program stored in the memory to realize the sepsis early risk prediction method.
It should be noted that the sequence of the 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 particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments.
Claims (3)
1. A method of predicting early risk of sepsis, the method comprising:
Obtaining clinical manifestations of different types of patients and numerical values of a plurality of sign detection types, wherein the different types of patients comprise sepsis patients and patients without sepsis;
according to the difference degree of different types of patients on clinical manifestations, combining the correlation degree between each sign detection type of sepsis patients and the clinical manifestations to obtain the performance correlation degree of each sign detection type;
Obtaining the importance of each physical sign detection type of the sepsis patient according to the numerical distribution difference condition of different types of patients on each physical sign detection type, the numerical distribution stability condition of the sepsis patient on each physical sign detection type and the performance association degree of each physical sign detection type;
Acquiring a sepsis risk model based on the importance and numerical distribution of the patient in each sign detection type;
the method for obtaining the performance association degree comprises the following steps:
clustering all clinical manifestations to obtain a characterization cluster;
For any sign detection type of sepsis patients, obtaining an associated index of the sign detection type in each characterization cluster according to the numerical value of the sign detection type of all sepsis patients in each characterization cluster and the relevant stability condition of the corresponding clinical manifestation;
according to the overall difference degree of clinical manifestations of patients suffering from sepsis and patients not suffering from sepsis in each characterization cluster, taking the overall difference degree as an expressive power index of each characterization cluster;
According to the association indexes of the sign detection type in each characterization cluster, combining the performance capability indexes of each characterization cluster to obtain the performance association degree of the sign detection type;
the method for acquiring the association index comprises the following steps:
In any characterization cluster, acquiring a sum vector of a numerical value of each sepsis patient in the sign detection type and a corresponding clinical manifestation, and taking the sum vector as a joint vector of each sepsis patient in the characterization cluster, wherein the joint vector corresponds to the sign detection type;
Calculating Euclidean distance between each joint vector and each other joint vector in the characterization cluster, and solving variance of all Euclidean distances to obtain distribution confusion of each joint vector; carrying out negative correlation mapping on the average value of the distribution confusion degree of all the joint vectors in the characterization cluster to obtain a joint stability index of the sign detection type in the characterization cluster;
Arranging the clinical manifestations of all sepsis patients in the characterization cluster according to the sequence of the corresponding numerical sequences to obtain the manifestation sequence of the characterization cluster;
Calculating the correlation between the numerical sequence and the expression sequence of the characterization cluster to obtain the correlation index of the sign detection type in the characterization cluster;
combining the combined stable index and the related index of the sign detection type in the characterization cluster to obtain the related index of the sign detection type in the characterization cluster;
the method for acquiring the expressive force index comprises the following steps:
For any one characterization cluster, obtaining the sum vector of clinical manifestations of all sepsis patients in the characterization cluster as a sepsis manifestation vector of the characterization cluster; obtaining the sum vector of clinical manifestations of all patients without sepsis in the characterization cluster, and taking the sum vector as the expression vector without sepsis of the characterization cluster;
taking the Euclidean distance between the sepsis expression vector of the characterization cluster and the sepsis expression vector without sepsis as the expression capability index of the characterization cluster;
The step of obtaining the performance association degree of the sign detection type according to the association index of the sign detection type in each characterization cluster and combining the performance capability index of each characterization cluster comprises the following steps:
Taking the product of the association index of the sign detection type in each characterization cluster and the expressive power index of the corresponding characterization cluster as the adjustment association degree of the sign detection type in each characterization cluster;
taking the average value of the adjustment association degree of the sign detection type in all the characterization cluster as the expression association degree of the sign detection type;
The method for acquiring the importance degree comprises the following steps:
For any sign detection type of a sepsis patient, obtaining a performance distinguishing index of the sign detection type according to the overall numerical difference degree of the sepsis patient and the non-sepsis patient on the sign detection type;
obtaining a performance stability index of the physical sign detection type according to the consistent degree of the numerical distribution of the sepsis patient on the physical sign detection type;
combining the performance distinguishing index, the performance stabilizing index and the performance association degree of the sign detection type to obtain the importance degree of the sign detection type;
the method for acquiring the performance differentiation index comprises the following steps:
Calculating the average value of the numerical values of all patients without sepsis in the physical sign detection type to obtain the diseased average value of the physical sign detection type;
Taking the difference between the diseased average value and the non-diseased average value of the sign detection type as a performance distinguishing index of the sign detection type;
the method for acquiring the performance stability index comprises the following steps:
And carrying out negative correlation mapping on the variances of the numerical values of all sepsis patients in the physical sign detection type to obtain the performance stability index of the physical sign detection type.
2. A method of predicting early risk of sepsis according to claim 1, wherein the method of obtaining a sepsis risk model comprises:
When the importance degree of the sign detection type is larger than a preset importance threshold value, the corresponding sign detection type is used as an evaluation type;
Taking the importance of the evaluation type of the patient as a weight, taking the value of the evaluation type as a training set to enter and exit into a logistic regression model for training to obtain a sepsis risk model after training, inputting the sepsis risk model as the value of the sign detection type of the patient, and outputting a risk evaluation result.
3. A sepsis early risk prediction system comprising a memory and a processor, wherein the processor executes a calculation program stored in the memory to implement a sepsis early risk prediction method according to any one of claims 1-2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411398140.9A CN118899088B (en) | 2024-10-09 | 2024-10-09 | A system and method for predicting early risk of sepsis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411398140.9A CN118899088B (en) | 2024-10-09 | 2024-10-09 | A system and method for predicting early risk of sepsis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118899088A CN118899088A (en) | 2024-11-05 |
| CN118899088B true CN118899088B (en) | 2025-01-21 |
Family
ID=93266731
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411398140.9A Active CN118899088B (en) | 2024-10-09 | 2024-10-09 | A system and method for predicting early risk of sepsis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118899088B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN120072273B (en) * | 2025-04-25 | 2025-10-03 | 榆林市中医医院 | Early warning method and system for rheumatic immune nephropathy based on data analysis |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111951975A (en) * | 2020-08-19 | 2020-11-17 | 哈尔滨工业大学 | An early warning method of sepsis based on deep learning model GPT-2 |
| CN116013466A (en) * | 2023-01-10 | 2023-04-25 | 中国人民解放军总医院第二医学中心 | A method and system for risk assessment of patients in perioperative period |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8615291B2 (en) * | 2009-03-13 | 2013-12-24 | National Institutes Of Health (Nih) | Method, system and computer program method for detection of pathological fluctuations of physiological signals to diagnose human illness |
| EP4573371A2 (en) * | 2022-08-19 | 2025-06-25 | The Johns Hopkins University | Diagnosis and treatment of acute kidney injury |
| CN117238510B (en) * | 2023-11-16 | 2024-02-06 | 天津中医药大学第二附属医院 | Sepsis prediction method and system based on deep learning |
| CN117976213A (en) * | 2024-02-04 | 2024-05-03 | 中国人民解放军联勤保障部队第九〇四医院 | Sepsis acute liver injury risk prediction method, device, equipment and storage medium |
| CN118379730B (en) * | 2024-04-18 | 2025-01-21 | 吉林大学 | A fine-grained classification method for skin basal cell carcinoma based on cluster analysis algorithm |
-
2024
- 2024-10-09 CN CN202411398140.9A patent/CN118899088B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111951975A (en) * | 2020-08-19 | 2020-11-17 | 哈尔滨工业大学 | An early warning method of sepsis based on deep learning model GPT-2 |
| CN116013466A (en) * | 2023-01-10 | 2023-04-25 | 中国人民解放军总医院第二医学中心 | A method and system for risk assessment of patients in perioperative period |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118899088A (en) | 2024-11-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110827993A (en) | Early death risk assessment model establishing method and device based on ensemble learning | |
| CN107194138B (en) | Fasting blood glucose prediction method based on physical examination data modeling | |
| CN118899088B (en) | A system and method for predicting early risk of sepsis | |
| CN114220540A (en) | Construction method and application of diabetic nephropathy risk prediction model | |
| CN112967803A (en) | Early mortality prediction method and system for emergency patients based on integrated model | |
| CN112183572A (en) | Method and device for generating prediction model for predicting pneumonia severity | |
| CN113057587A (en) | Disease early warning method and device, electronic equipment and storage medium | |
| CN118762849A (en) | An intelligent processing method and system for urology diagnosis and treatment data | |
| CN115568845A (en) | A prediction method, kit and system for acute respiratory distress syndrome after liver transplantation | |
| CN118658629A (en) | Coronary heart disease risk auxiliary evaluation system based on intelligent medical treatment | |
| CN116864062A (en) | Health physical examination report data analysis management system based on Internet | |
| CN120019446A (en) | Systems and methods for screening and predicting sepsis | |
| CN117174321A (en) | Thrombolysis prediction model construction method based on machine learning | |
| CN119069002B (en) | A method, device, medium and product for predicting venous thromboembolism based on genetic and non-genetic characteristics | |
| Eskandari et al. | Detection of sepsis using biomarkers based on machine learning | |
| CN119694579A (en) | Method and device for constructing female osteoporosis prediction model based on physical examination data | |
| Sang et al. | Study on survival prediction of patients with heart failure based on support vector machine algorithm | |
| Hassan et al. | Prediction of Chronic Diseases using Machine Learning Classifiers | |
| CN117373659A (en) | Diabetes prediction model and construction method and application thereof | |
| CN117976214A (en) | Method, device, equipment and storage medium for predicting 28-day death risk of senile sepsis | |
| CN117116475A (en) | Method, system, terminal and storage medium for predicting risk of ischemic cerebral apoplexy | |
| CN113035352B (en) | Early warning method of diabetic retinopathy based on BP neural network | |
| Choi et al. | Artificial Intelligence for Predicting Mortality Due to Sepsis | |
| Ghias et al. | Using Machine Learning Algorithms to predict sepsis and its stages in ICU patients | |
| Darwiche et al. | Improving septic shock prediction with AdaBoost and cox regression model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |