A method and system for predicting the rationality of antibiotic use

Abstract

The present invention relates to a method and system for predicting the rationality of antibiotic use, the method comprising: obtaining structured and unstructured relevant data of patients who have used antibiotics in the past; performing natural language processing on the unstructured data of the patients, including word segmentation processing and information extraction; performing centralization, discretization processing, and data dimension reduction processing on the structured data and the unstructured data after natural language processing to form a data set; inputting the data set into a random forest algorithm to train the random forest algorithm; inputting the relevant data of new patients into the trained random forest algorithm, and the output result is the judgment result of whether the antibiotics are reasonable. Based on the real data of previous patients, the present invention provides a more reasonable antibiotic treatment plan for new patients, accumulates treatment experience for doctors, improves the efficiency of clinical diagnosis and treatment, and brings greater benefits to patients.

Description

Technical Field

[0001] The present application relates to the field of medical technology, and in particular to a method and system for predicting the rationality of antibiotic use. Background Art

[0002] Antibiotics are the main drugs for treating infectious diseases. The correct use of antibiotics can achieve a good outcome and prognosis for infectious diseases. With the popularization and extensive use of antibiotics in clinical practice, in recent years, due to the irrational and incorrect use of antibiotics, adverse reactions such as drug-resistant strains, toxic reactions, allergic reactions and superinfections have increased, bringing additional physiological and economic burdens to patients and hindering the rational allocation and use of medical resources.

[0003] How to guide the effective and rational use of antibiotics in clinical practice is an urgent problem to be solved. Guiding clinical medical workers to use reasonable and effective antibiotics for patients can enable patients to receive correct treatment for infectious diseases, reduce the physiological and economic burden on patients, and enable the rational allocation and use of medical resources.

[0004] Therefore, providing a method and system for predicting the rationality of antibiotic use is the main problem that needs to be solved at present. Summary of the invention

[0005] The present application provides a method and system for predicting the rationality of antibiotic use to guide clinical medical workers to use reasonable and effective antibiotics for patients, so that patients' infectious diseases can be correctly treated, the physiological and economic burdens of patients can be reduced, and medical resources can be reasonably allocated and used.

[0006] The technical solutions adopted in this application are as follows:

[0007] In a first aspect, the present invention provides a method for predicting the rationality of antibiotic use, comprising:

[0008] Obtain structured and unstructured relevant data of patients with previous antibiotic use, the structured relevant data including: gender, age, smoking history, hypertension, coronary heart disease, white blood cell count, platelet count, procalcitonin, C-reactive protein, main diagnosis name, fever, body temperature, bowel movement frequency, previous surgery name, previous antibiotic name, concomitant drug name, genetic testing data and clinical scoring scale data, the unstructured data including: basic information, chief complaint, current medical history, past medical history, ward rounds and X-ray examination report;

[0009] Performing natural language processing on the unstructured data, including word segmentation and information extraction;

[0010] Centralize, discretize, and reduce the dimension of the structured data and the unstructured data after natural language processing to form a data set;

[0011] Inputting the data set into a random forest algorithm to train the random forest algorithm;

[0012] The relevant data of the new patient is input into the trained random forest algorithm, and the output result is the judgment result whether the antibiotic is reasonable.

[0013] Furthermore, the relevant data can be obtained from hospital information systems, laboratory information systems, radiology department information systems and electronic medical records.

[0014] Furthermore, the structured related data specifically include: gender, age, smoking history, hypertension, coronary heart disease, white blood cell count, platelet count, procalcitonin, C-reactive protein, main diagnosis name, fever, body temperature, bowel movement frequency, previous surgery name, previous antibiotic name, concomitant drug name, genetic test data and clinical scoring scale data;

[0015] The unstructured data includes: basic information, chief complaint, current medical history, past medical history, ward rounds records and X-ray examination reports.

[0016] Furthermore, the word segmentation process includes:

[0017] Process the symptoms in the patient's complaint into the form of symptom name / symptom;

[0018] Process the medication information in the patient's current medical history into the form of drug name / drug;

[0019] Words other than symptoms and medication information were processed into text / part-of-speech forms.

[0020] Furthermore, the information extraction includes:

[0021] Input the word segmentation processing result into a regular expression, and the obtained output result is in the form of structured related data;

[0022] The regular expression presets a matching pattern to be extracted, and the matching pattern is: ,.*\ / symptom and ,.*\ / drug;

[0023] The output structured related data include:,.*\ / symptom names corresponding to symptoms and,.*\ / drug names corresponding to drugs.

[0024] Furthermore, the information extraction includes:

[0025] Input the word segmentation processing result into a regular expression, and the obtained output result is in the form of structured related data;

[0026] The regular expression is preset with words to be extracted, and the words are: symptoms and medicines;

[0027] The output structured related data includes: symptom names corresponding to symptoms and drug names corresponding to drugs.

[0028] Furthermore, the centralization processing includes: centralization processing of platelet count; the discretization processing includes: discretization processing of the patient's age, body temperature, and C-reactive protein; the data dimension reduction processing includes: data dimension reduction processing of the patient's genetic test data and clinical scoring scale data through principal component extraction method.

[0029] In an achievable embodiment, the data set is input into a random forest algorithm, and training is performed on the data set, including:

[0030] Create multiple decision trees, and use a random sampling method with replacement to extract a preset number of samples from the data set for each decision tree, and train each decision tree separately;

[0031] The out-of-bag error rate is used to evaluate the performance of the decision tree. The out-of-bag error rate is calculated as follows: for each sample, the classification of the tree as an out-of-bag sample is calculated, and then a simple majority vote is used as the classification result of the sample. Finally, the ratio of the number of misclassified samples to the total number of samples is used as the out-of-bag error rate of the random forest.

[0032] If the out-of-bag error rate does not meet expectations, the parameters of the random forest algorithm are reset, including the number of samples, the number of sub-classifiers, and the node depth, and the out-of-bag error rate is recalculated until the out-of-bag error rate meets the expected standard.

[0033] In an achievable embodiment, the relevant data of a new patient is input into the trained random forest algorithm, and the output result is the judgment result of whether the use of antibiotics is reasonable, including:

[0034] Input the relevant data of the new patient into the trained random forest algorithm;

[0035] If the prediction result of the majority of decision trees is that the use of antibiotics is reasonable, the majority vote of the decision trees is adopted, and the output result is that the use of antibiotics is reasonable, and the majority is at least 80% of the number of decision trees;

[0036] If the prediction result of most decision trees is that the use of antibiotics is unreasonable, the majority vote of the decision trees is adopted and the output result is that the use of antibiotics is unreasonable.

[0037] In a second aspect, the present invention further provides a system for predicting the rationality of antibiotic use, which is used to implement the above method, comprising:

[0038] A data extraction module, the data extraction module is used to obtain structured and unstructured related data of patients using antibiotics, the structured related data includes: gender, age, smoking history, hypertension, coronary heart disease, white blood cell count, platelet count, procalcitonin, C-reactive protein, main diagnosis name, fever, body temperature, bowel movement frequency, previous surgery name, previous antibiotic name, concomitant drug name, genetic testing data and clinical scoring scale data, the unstructured data includes: basic information, chief complaint, current medical history, past medical history, ward rounds and X-ray examination report;

[0039] A data cleaning module, wherein the data cleaning module is used to perform natural language processing on the unstructured data, including word segmentation processing and information extraction;

[0040] A feature engineering module, wherein the feature engineering module is used to perform centralization, discretization, and data dimension reduction on the structured data and the unstructured data after natural language processing to form a data set;

[0041] A model training module, wherein the model training module is used to input the data set into a random forest algorithm to train the random forest algorithm;

[0042] The medication rationality prompt module is used to input the relevant data of new patients into the trained random forest algorithm, and the output result is the judgment result of whether the antibiotics are reasonable.

[0043] In a third aspect, the present invention further provides a medical device comprising the system as described above.

[0044] The beneficial effects of adopting the technical solution of this application are as follows:

[0045] The invention discloses a method for predicting the rationality of antibiotic use, comprising obtaining structured and unstructured relevant data of patients who have used antibiotics in the past; performing natural language processing on the unstructured data of the patients, including word segmentation processing and information extraction; performing centralization, discretization processing, and data dimension reduction processing on the structured data and the unstructured data after natural language processing to form a data set; inputting the data set into a random forest algorithm to train the random forest algorithm; inputting relevant data of new patients into the trained random forest algorithm, and outputting a result, which is a judgment result of whether the antibiotics are reasonable.

[0046] The present invention can provide new patients with more reasonable antibiotic treatment plans based on the real data of previous patients, accumulate treatment experience for doctors, improve the efficiency of clinical diagnosis and treatment, and bring greater benefits to patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0048] Figure 1 This is a flowchart of a method for predicting the rationality of antibiotic use;

[0049] Figure 2 To indicate participle Figure 1 ;

[0050] Figure 3 To indicate participle Figure 2 ;

[0051] Figure 4 Information extraction instructions Figure 1 ;

[0052] Figure 5 Information extraction instructions Figure 2 ;

[0053] Figure 6 Schematic diagram of a system for predicting the rationality of antibiotic use. DETAILED DESCRIPTION

[0054] The following embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following embodiments do not represent all implementations consistent with the present application. They are only examples of systems and methods consistent with some aspects of the present application as detailed in the claims.

[0055] like Figure 1 As shown, as a first aspect, the present invention discloses a method for predicting the rationality of antibiotic use, comprising:

[0056] S01: Obtain structured and unstructured data related to several patients with previous antibiotic use.

[0057] Among them, relevant data can be obtained from HIS (hospital information system), LIS (laboratory information system), RIS (radiology information system) and EMR (electronic medical record).

[0058] Further, the structured related data are shown in Tables 1 and 2 below:

[0059]

[0060]

[0061] Furthermore, the unstructured related data includes the patient's basic information, chief complaint, current medical history, past medical history, ward rounds record and X-ray examination report, as shown in Table 3, Table 4 and Table 5:

[0062]

[0063]

[0064] S02: Performing natural language processing on the unstructured data, including word segmentation and information extraction.

[0065] The natural language processing of the acquired unstructured data is mainly to extract the data related to the patient's disease information from the data recorded in text form. First, the word segmentation is performed, that is:

[0066] The patient's symptoms are processed into the form of symptom name / symptom; the patient's medication information is processed into the form of drug name / drug; and words other than symptoms and medication information are processed into the form of text / part of speech.

[0067] like Figure 2 (Participle indication Figure 1 ), in the example, the symptom "nausea and vomiting" is segmented into "nausea / a, vomiting / n", where the word is before " / ", and the part of speech is after " / ", such as "nausea / a" means that the text "nausea" is a word and its part of speech is "a", which means an adjective; in the example, "Huoxiang Zhengqi Water" is segmented into "Huoxiang / n, Zhengqi Water / n", but "Huoxiang Zhengqi Water" is a complete word, indicating the name of a medicine; in order to solve the above-mentioned word segmentation granularity problem, a dictionary is used to collect and organize medical vocabulary from the hospital database and manually annotate the part of speech as the basis for word segmentation. The word record format in the dictionary is "the number of occurrences of the word's custom part of speech". An example of the dictionary is shown in Table 6:

[0068]

[0069] Use a custom dictionary as the basis for word segmentation. Figure 2 After the example text is segmented, it can be achieved as follows Figure 3 (Participle indication Figure 2 ) shows the effect:

[0070] After customizing the part of speech, symptom words such as "nausea" are segmented into "nausea / symptoms", and words representing medicines such as "Huoxiang Zhengqi Water" are segmented into "Huoxiang Zhengqi Water / Medicine", which is convenient for subsequent information extraction; the word segmenter used in the above example is the conditional random field (CRF) word segmenter. In addition, the word segmentation processing technology can also use different word segmenters, such as the perceptron word segmenter.

[0071] After word segmentation is completed, the next step is information extraction, the purpose of which is to extract useful information from unstructured data such as patient symptoms, patient medication use, and other data.

[0072] An information extraction technology based on regular expression (Regular Expression) rules can be used: the word segmentation processing result is input into a regular expression, and the output result is in the form of structured related data; the regular expression presets a matching pattern to be extracted, and the matching pattern is: ",.*\ / symptom" and ",.*\ / drug"; the output structured related data includes: the symptom name corresponding to ",.*\ / symptom" and the drug name corresponding to ",.*\ / "drug.

[0073] The information extraction technology based on regular expression rules, its implementation example and pseudo code are as follows Figure 4 (Information extraction diagram Figure 1 ), where: for the text data after word segmentation, if you want to extract the patient's symptom information, in the text after word segmentation, the words representing the symptom information meet the characteristics of ", words / symptoms", then edit the matching pattern ",.*\ / symptoms", the matching pattern means: ".*" means matching any character, ",.*" means matching characters starting with "," ".*\ / symptoms" means matching characters ending with "\ / symptoms", so ",.*\ / symptoms" means matching text that starts with ", " and ends with "\ / symptoms" and contains any characters in the middle, then the words that meet the matching pattern ",.*\ / symptoms" are recognized as information describing the patient's symptoms, and the words are extracted, such as the text after word segmentation ", nausea / symptoms"; similarly, the words that meet the matching pattern ",.*\ / drugs" are recognized as information describing the patient's use of drugs, and the words are extracted, and the extracted information is organized into structured data {symptoms: ["nausea", "vomiting"], drugs: "Huoxiang Zhengqi Water"}.

[0074] Information extraction technology based on part-of-speech rules can also be used: the word segmentation processing result is input into a regular expression, and the output result is in the form of structured related data; the regular expression presets the words to be extracted, and the words are: symptoms and medicines; the output structured related data includes: symptom names corresponding to symptoms and drug names corresponding to drugs.

[0075] Information extraction technology based on part-of-speech rules, its implementation example and pseudocode Figure 5 (Information extraction diagram Figure 2) as shown in: For the text data after word segmentation, if the part of speech of the word is "symptom", the text of the word is extracted as the patient's symptom; if the part of speech of the word is "drug", the text of the word is extracted as the drug used by the patient; the extracted information is organized into structured data {symptoms: ["nausea", "vomiting"], drug: "Huoxiang Zhengqi Water"}.

[0076] After completing the natural language processing of the unstructured data, the unstructured data that has been structured by the patient is merged with the structured data collected at the beginning through the patient ID, and after the merger, a complete data set of patients using antibiotics is obtained.

[0077] S03: Centralize, discretize, and reduce the dimension of the structured data and the unstructured data after natural language processing to form a data set.

[0078] Since the structured data obtained in S02 and the unstructured data after natural language processing have the characteristics of large distribution dispersion of some features and high dimension of some features, in order to facilitate the subsequent training of machine learning algorithms and reduce the complexity of the algorithms, feature engineering is required, mainly including discretization, centering, data dimensionality reduction and other processing of features.

[0079] For age and body temperature that need to be discretized, then:

[0080] The age of the patients is discretized, that is, the age of the patients is divided into different preset ranges according to size; for example, the age feature is continuously distributed, but the use of antibiotics has the characteristic that different types of antibiotics are suitable for different age groups, so the age can be discretized, such as the age of patient 1 is 23 years old, which is 18~24 years old after conversion, and the age of patient 2 is 6 years old, which is 0~6 years old after conversion.

[0081] The patient's body temperature condition is discretized, that is, the patient's body temperature is divided into normal if it is within the preset range, and abnormal if it is not within the preset range; for example, the body temperature feature is continuously distributed, but the body temperature within a certain range does not affect the patient's choice of antibiotics, and the body temperature is generally divided into only two conditions, "normal" and "abnormal", so the continuous body temperature data can be discretized, such as the age of patient 1 is 36.3℃, which is "normal" after conversion, and the age of patient 2 is 38.5℃, which is "abnormal" after conversion.

[0082] For platelet counts that need to be centralized, for example, platelet counts are continuous data with a large degree of variation, and can be centralized into continuous data with uniform distribution. The conversion method is:

[0083] ;

[0084] in, is the converted platelet count value, is the platelet count value before transformation, is the minimum platelet count value before conversion. It is the maximum platelet count value before conversion.

[0085] Assuming that the minimum platelet count value is 125 and the maximum platelet count value is 350, the platelet count of case 1 was 323 before conversion and 0.88 after conversion, and the platelet count of case 2 was 216 before conversion and 0.404 after conversion.

[0086] For C-reactive protein that needs to be centralized, for example, C-reactive protein is continuous data that obeys a normal distribution, and it can be converted into data that obeys a standard normal distribution by:

[0087] ;

[0088] in, is the converted C-reactive protein value, is the C-reactive protein value before conversion, μ is the mean value of C-reactive protein, and σ is the standard deviation of C-reactive protein.

[0089] Assuming that the mean of C-reactive protein calculated from the data set is 1 and the standard deviation is 0.5, the C-reactive protein of case 1 was 1.06 before conversion and 0.792 after conversion, and the C-reactive protein of case 2 was 0.89 before conversion and 0.779 after conversion.

[0090] For C-reactive protein that needs dimensionality reduction, and for patients' genetic testing data and clinical scoring scale data, the dimensions of these data can be as high as dozens or even hundreds or thousands. High-dimensional data is not conducive to algorithm training, so these data can be reduced in dimension, mainly using the principal component extraction method.

[0091] For example, for the patient's symptom self-assessment scale SCL90 data, the data contains 90 dimensions, but there are certain connections between the dimensions, so principal component analysis is used to extract them into 10 dimensions. The extracted 10 dimensions can contain most of the information of the original 90 dimensions but the dimensionality is lower.

[0092] S04: Inputting the data set into a random forest algorithm to train the random forest algorithm.

[0093] An integrated classification algorithm is used to model and predict the antibiotic use rationality data, and the algorithm used is random forest.

[0094] The specific steps include:

[0095] Model fitting: Create multiple decision trees, and use a random sampling method with replacement to extract a preset number of samples from the data set obtained in S03 for each decision tree, and train each decision tree separately;

[0096] The feature data set is trained using the random forest algorithm with decision tree as a subclassifier, and parameters such as the number of samples, the number of subclassifiers, and the node depth are continuously adjusted to ensure that the model achieves the expected performance.

[0097] If the decision tree subclassifier used is the CD3 algorithm, the decision tree as a random forest subclassifier performs node splitting based on information entropy, which is defined as:

[0098] ;

[0099] In the process of creating a decision tree, each node split should maximize the decrease in entropy value until all the data in the node belong to the same category.

[0100] Model evaluation: The out-of-bag error rate is used to evaluate the performance of the decision tree. The out-of-bag error rate is calculated as follows: for each sample, the classification of the tree as an out-of-bag sample is calculated, and then a simple majority vote is used as the classification result of the sample. Finally, the ratio of the number of misclassified samples to the total number of samples is used as the out-of-bag error rate of the random forest.

[0101] Model selection: If the out-of-bag error rate does not meet expectations, the parameters of the random forest algorithm are reset, including the number of samples, the number of sub-classifiers, and the node depth, and the out-of-bag error rate is recalculated until the out-of-bag error rate meets the expected standard.

[0102] Exemplarily, after natural language processing and feature engineering of S02 and S03, a total of 10,000 antibiotic usage data are obtained. The data have been labeled with the rationality of antibiotic use. These data are used to train the random forest algorithm. The algorithm parameters are set to create 100 sub-classifiers. In this case, 100 decision trees are created. The training samples of each sub-classifier are 1,000. The out-of-bag error rate of the algorithm is 2% after performance evaluation, that is, only 2 predictions are wrong out of 100 samples. The model meets expectations and can be used to predict the rationality of antibiotic use for new patients.

[0103] S05: Input the relevant data of the new patient into the trained random forest algorithm, and the output result is the judgment result of whether the antibiotics are reasonable.

[0104] The specific steps include:

[0105] Input the relevant data of the new patient into the trained random forest algorithm;

[0106] The algorithm will automatically predict the appropriateness of antibiotic use for new patients by:

[0107] If the prediction result of the majority of decision trees is that the use of antibiotics is reasonable, the majority vote of the decision trees is adopted, and the output result is that the use of antibiotics is reasonable, and the majority is at least 80% of the number of decision trees;

[0108] If the prediction result of most decision trees is that the use of antibiotics is unreasonable, the majority vote of the decision trees is adopted and the output result is that the use of antibiotics is unreasonable.

[0109] Exemplarily, the random forest algorithm has a total of 100 sub-classifiers. After receiving new patient data, 98 sub-classifiers predict that antibiotic use is reasonable, and 2 sub-classifiers predict that antibiotic use is unreasonable. The algorithm uses the majority vote of the sub-classifiers and finally predicts that the antibiotic use of the new patient is reasonable.

[0110] The method of the present invention uses a dictionary based on database collation to segment unstructured data, uses rules based on part of speech to extract information, and then uses a random forest algorithm to predict the rationality of antibiotic use. This enables the present invention to provide new patients with more reasonable antibiotic treatment plans based on real data of previous patients, accumulate treatment experience for doctors, improve the efficiency of clinical diagnosis and treatment, and bring greater benefits to patients.

[0111] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0112] As a second aspect, the present invention further discloses a system for predicting the rationality of antibiotic use, which is used to implement the above method, comprising:

[0113] A data extraction module, the data extraction module is used to obtain structured and unstructured related data of patients using antibiotics, the structured related data includes: gender, age, smoking history, hypertension, coronary heart disease, white blood cell count, platelet count, procalcitonin, C-reactive protein, main diagnosis name, fever, body temperature, bowel movement frequency, previous surgery name, previous antibiotic name, concomitant drug name, genetic testing data and clinical scoring scale data, the unstructured data includes: basic information, chief complaint, current medical history, past medical history, ward rounds and X-ray examination report;

[0114] A data cleaning module, wherein the data cleaning module is used to perform natural language processing on the unstructured data, including word segmentation processing and information extraction;

[0115] A feature engineering module, wherein the feature engineering module is used to perform centralization, discretization, and data dimension reduction on the structured data and the unstructured data after natural language processing to form a data set;

[0116] A model training module, wherein the model training module is used to input the data set into a random forest algorithm to train the random forest algorithm;

[0117] The module for prompting rational medication is used to input relevant data of new patients into the trained random forest algorithm, and the output result is the judgment result of whether the antibiotics are reasonable (medication regimen recommendation).

[0118] For specific limitations on the antibiotic medication rationality prediction system, please refer to the limitations on the antibiotic medication rationality prediction method above, and no further details will be given here. In addition, each module in the above-mentioned antibiotic medication rationality prediction system can be implemented in whole or in part through software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0119] In a third aspect, the present invention further provides a medical device comprising the system as described above.

[0120] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the general concept of this application and do not constitute a limitation on the protection scope of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without creative work belong to the protection scope of this application.

Claims

1. A method for predicting the rationality of antibiotic use, characterized in that: include: Obtain structured and unstructured related data of patients with previous antibiotic use, the structured related data include: gender, age, smoking history, hypertension, coronary heart disease, white blood cell count, platelet count, procalcitonin, C-reactive protein, main diagnosis name, fever, body temperature, bowel movement frequency, previous surgery name, previous antibiotic name, concomitant drug name, genetic testing data and clinical scoring scale data, the unstructured data include: basic information, chief complaint, current medical history, past medical history, ward rounds and X-ray examination report; Performing natural language processing on the unstructured data, including word segmentation and information extraction; Centralize, discretize, and reduce the dimension of the structured data and the unstructured data after natural language processing to form a data set; The centralization includes: centralizing platelet count; For the platelet count in the centralized processing, the platelet count is continuous data with a large degree of variation. It is centralized into continuous data with uniform distribution. The conversion method is: ； in, is the converted platelet count value, is the platelet count value before transformation, is the minimum platelet count value before conversion. It is the maximum value of platelet count before transformation; The centralization also includes: centralizing the C-reactive protein; For the centralized C-reactive protein, which is continuous data that obeys the normal distribution, it is converted into data that obeys the standard normal distribution by the following conversion method: ； in, is the converted C-reactive protein value, is the C-reactive protein value before conversion, μ is the mean value of C-reactive protein, and σ is the standard deviation of C-reactive protein; The discretization process includes: discretization of the patient's age, body temperature, and C-reactive protein; The data dimension reduction processing includes: performing data dimension reduction processing on the patient's gene detection data and clinical scoring scale data by principal component extraction; Inputting the data set into a random forest algorithm to train the data set; The relevant data of the new patient is input into the trained random forest algorithm, and the output result is the judgment result whether the antibiotic is reasonable; The step of inputting the data set into a random forest algorithm and training the data set includes: Create multiple decision trees, and use a random sampling method with replacement to extract a preset number of samples from the data set for each decision tree, and train each decision tree separately; The out-of-bag error rate is used to evaluate the performance of the decision tree. The out-of-bag error rate is calculated as follows: for each sample, the classification of the tree as an out-of-bag sample is calculated, and then a simple majority vote is used as the classification result of the sample. Finally, the ratio of the number of misclassified samples to the total number of samples is used as the out-of-bag error rate of the random forest. If the out-of-bag error rate does not meet expectations, the parameters of the random forest algorithm are reset, including the number of samples, the number of sub-classifiers, and the node depth, and the out-of-bag error rate is recalculated until the out-of-bag error rate meets the expected standard; The relevant data of the new patient is input into the trained random forest algorithm, and the output result is the judgment result of whether the use of antibiotics is reasonable, including: Input the relevant data of the new patient into the trained random forest algorithm; If the prediction result of the majority of decision trees is that the use of antibiotics is reasonable, the majority vote of the decision trees is adopted, and the output result is that the use of antibiotics is reasonable, and the majority is at least 80% of the number of decision trees; If the prediction result of most decision trees is that the use of antibiotics is unreasonable, the majority vote of the decision trees is adopted and the output result is that the use of antibiotics is unreasonable.

2. The method for predicting the rationality of antibiotic use according to claim 1, characterized in that: The relevant data are obtained from the hospital information system, laboratory information system, radiology department information system and electronic medical records.

3. The method for predicting the rationality of antibiotic use according to claim 1 or 2, characterized in that: The word segmentation process includes: Process the symptoms in the patient's complaint into the form of symptom name / symptom; Process the medication information in the patient's current medical history into the form of drug name / drug; Words other than symptoms and medication information were processed into text / part-of-speech forms.

4. The method for predicting the rationality of antibiotic use according to claim 3, characterized in that: The information extraction includes: The symptom name / symptom form, the drug name / drug form, and the text / part of speech form processed by the word segmentation are input into a regular expression, and the output result obtained is in the form of structured related data; The regular expression presets a matching pattern to be extracted, and the matching pattern is: ",.*\ / symptom" and ",.*\ / drug"; The output structured related data include: ".*\ / symptoms" corresponding to the symptom names and ".*\ / drugs" corresponding to the drug names.

5. The method for predicting the rationality of antibiotic use according to claim 3, characterized in that: The information extraction includes: The symptom name / symptom form, the drug name / drug form, and the text / part of speech form processed by the word segmentation are input into a regular expression, and the output result obtained is in the form of structured related data; The regular expression is preset with words to be extracted, and the words are: symptoms and medicines; The output structured related data includes: symptom names corresponding to symptoms and drug names corresponding to drugs.

6. A system for predicting the rationality of antibiotic use, used to implement the method described in any one of claims 1 to 5, characterized in that: include: A data extraction module, the data extraction module is used to obtain structured and unstructured related data of patients using antibiotics, the structured related data includes: gender, age, smoking history, hypertension, coronary heart disease, white blood cell count, platelet count, procalcitonin, C-reactive protein, main diagnosis name, fever, body temperature, bowel movement frequency, previous surgery name, previous antibiotic name, concomitant drug name, genetic testing data and clinical scoring scale data, the unstructured data includes: basic information, chief complaint, current medical history, past medical history, ward rounds and X-ray examination report; A data cleaning module, wherein the data cleaning module is used to perform natural language processing on the unstructured data, including word segmentation processing and information extraction; A feature engineering module, wherein the feature engineering module is used to perform centralization, discretization, and data dimension reduction on the structured data and the unstructured data after natural language processing to form a data set; The centralization includes: centralizing platelet count; For the platelet count in the centralized processing, the platelet count is continuous data with a large degree of variation. It is centralized into continuous data with uniform distribution. The conversion method is: ； in, is the converted platelet count value, is the platelet count value before transformation, is the minimum platelet count value before conversion. It is the maximum value of platelet count before transformation; The centralization also includes: centralizing the C-reactive protein; For the centralized C-reactive protein, which is continuous data that obeys the normal distribution, it is converted into data that obeys the standard normal distribution by the following conversion method: ； in, is the converted C-reactive protein value, is the C-reactive protein value before conversion, μ is the mean value of C-reactive protein, and σ is the standard deviation of C-reactive protein; The discretization process includes: discretization of the patient's age, body temperature, and C-reactive protein; The data dimension reduction processing includes: performing data dimension reduction processing on the patient's gene detection data and clinical scoring scale data by principal component extraction; A model training module, wherein the model training module is used to input the data set into a random forest algorithm to train the random forest algorithm; A medication rationality prompting module, which is used to input relevant data of new patients into the trained random forest algorithm, and the output result is the judgment result of whether the antibiotics are reasonable; Wherein, the model training module is further configured as follows: Create multiple decision trees, and use a random sampling method with replacement to extract a preset number of samples from the data set for each decision tree, and train each decision tree separately; The out-of-bag error rate is used to evaluate the performance of the decision tree. The out-of-bag error rate is calculated as follows: for each sample, the classification of the tree as an out-of-bag sample is calculated, and then a simple majority vote is used as the classification result of the sample. Finally, the ratio of the number of misclassified samples to the total number of samples is used as the out-of-bag error rate of the random forest. If the out-of-bag error rate does not meet expectations, the parameters of the random forest algorithm are reset, including the number of samples, the number of sub-classifiers, and the node depth, and the out-of-bag error rate is recalculated until the out-of-bag error rate meets the expected standard; The medication rationality prompt module is further configured as follows: Input the relevant data of the new patient into the trained random forest algorithm; If the prediction result of the majority of decision trees is that the use of antibiotics is reasonable, the majority vote of the decision trees is adopted, and the output result is that the use of antibiotics is reasonable, and the majority is at least 80% of the number of decision trees; If the prediction result of most decision trees is that the use of antibiotics is unreasonable, the majority vote of the decision trees is adopted and the output result is that the use of antibiotics is unreasonable.

7. A medical device, characterized in that: Comprising a system as claimed in claim 6.

Images