Electronic medical record-based infectious disease outbreak risk prediction method
By extracting and cleaning up symptom information related to infectious disease from electronic medical records, using the XGBoost model for prediction, and combining different symptom combinations, the problem of excessive application scope in the existing technology is solved, and the risk of infectious disease outbreaks is achieved.
Patent Information
- Application Number
- CN202410829515.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-25
- Publication Date
- 2024-11-12
AI Technical Summary
The existing infectious disease outbreak risk prediction methods rely on separate electronic medical record extraction technology and symptom monitoring and early warning technology. The application scope is too vague and broad, making it difficult to achieve specific and accurate predictions.
By extracting case information of target infectious disease-related symptoms from electronic medical records, forming initial and predicted data sets, using XGBoost model for training and prediction, combining different symptoms combinations to achieve more accurate prediction, and setting preset thresholds to judge infection The disease outbreak.
It realizes the risk prediction of infectious disease outbreaks based on electronic medical records, improves the accuracy and specificity of the prediction, has a wide range of application, and can effectively predict the risk of outbreaks of new infectious diseases.
Smart Images

Figure CN118942725A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of infectious disease outbreak risk prediction methods, and specifically to an infectious disease outbreak risk prediction method based on electronic medical records. Background Art
[0002] Most infectious diseases are discovered through symptomatic patients. At present, many medical institutions have achieved widespread use of electronic medical record systems. These systems collect a large amount of patient data, including information such as medical history, chief complaints, diagnosis, treatment and laboratory results. These data provide a rich basis for predicting the risk of infectious disease outbreaks. Public health agencies in some regions and countries use electronic medical record data for infectious disease monitoring and outbreak prediction. For example, the Centers for Disease Control and Prevention (CDC) in the United States and the European Center for Disease Prevention and Control (ECDC) rely on electronic medical record data for monitoring. Therefore, based on a mature hospital information management system, by monitoring and analyzing the relevant symptoms that may be caused by infectious diseases, it can be used to predict infectious disease outbreaks. This involves three links: electronic medical record information extraction, data cleaning, statistics, and infectious disease outbreak risk prediction. At present, electronic medical record extraction technology and symptom-based monitoring and early warning technology are separate applications, and the scope of application is too vague and broad. Summary of the invention
[0003] The technical problem to be solved by the present invention is to provide a method for predicting the risk of infectious disease outbreaks based on electronic medical records.
[0004] The specific scheme of the present invention is as follows:
[0005] A method for predicting the risk of infectious disease outbreaks based on electronic medical records, comprising the following steps:
[0006] Step 1: Extract case information containing symptoms related to the target infectious disease from the electronic medical records of past medical institutions to obtain the initial training data set;
[0007] Step 2: Clean, assign values, and count the initial training data set to obtain the number of visits for each symptom on each visit date as a feature training data set, and collect the actual number of reported cases on each visit date as a label;
[0008] Step 3: Input the feature training data set into the XGBoost model for training. The model outputs the predicted number of cases. By minimizing the error between the number of cases predicted by the model and the number of actual reported cases, the trained target infectious disease outbreak risk prediction XGBoost model is obtained.
[0009] Step 4: Extract case information containing symptoms related to the target infectious disease from the chief complaint in the electronic medical records of medical institutions on that day to obtain the initial dataset to be predicted; perform data cleaning, value assignment, and statistics on the initial dataset to be predicted to obtain the number of visits for each symptom on that day as the feature dataset.
[0010] Step 5: Input the feature data set into the trained target infectious disease outbreak risk prediction XGBoost model, output the case data prediction result of the day, and compare the prediction result with the preset threshold. If it is greater than the preset threshold, it is judged that the target infectious disease has broken out; otherwise, it is judged that the target infectious disease has not broken out.
[0011] Furthermore, in step 1, infectious disease-related symptoms are determined based on the symptoms that may be caused by the disease, and a variety of symptom combinations are obtained by arranging and combining the symptoms. In step 2, feature training data sets for different symptom combinations are obtained respectively. In step 3, XGBoost model training results corresponding to the feature training data sets for different symptom combinations are obtained, and the XGBoost model with the best effect is selected as the final training model.
[0012] Furthermore, the error in step 3 includes the root mean square error and the mean absolute error.
[0013] Furthermore, the step 2 specifically includes the following steps:
[0014] Clean up duplicate medical information, assign 0 to the medical information that contains the corresponding symptoms, otherwise assign 1.
[0015] Furthermore, in step 2, each feature training data includes the date of consultation, the number of consultations corresponding to each symptom in the symptom combination, and the actual number of reported cases on that day.
[0016] The beneficial effect of the present invention is that the method in this patent can be used when predicting the outbreak of a new and sudden infectious disease based on the electronic medical records of medical institutions. First, case information is extracted from the chief complaint of the electronic medical record according to the determined symptom list. Then the extracted data is cleaned, assigned, and counted to form a data set. Finally, the data set is input into the prediction model to achieve the prediction of infectious disease outbreaks. The present invention clarifies the relevant symptoms in infectious disease symptom monitoring, uses different symptom combinations to achieve more accurate predictions, and combines electronic medical record extraction with infectious disease outbreak risk prediction to form a complete process from electronic medical record extraction to prediction. The existing technical scope of infectious disease prediction based on symptom monitoring is too broad, and it is difficult to apply it specifically. The present invention uses mature medical institution information systems to extract information such as symptoms in electronic medical records, and models them through machine learning methods to predict infectious disease outbreaks. Specific symptoms can be adjusted and combined according to the infectious disease to be predicted, and the outbreak threshold can also be determined according to actual conditions, with a wide range of applications. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION
[0018] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than the entire embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0019] like Figure 1 As shown, a method for predicting the risk of infectious disease outbreaks based on electronic medical records includes the following steps:
[0020] Step 1. Extract case information of patients whose chief complaint contains symptoms related to infectious diseases (fever, cough, sore throat, nausea / vomiting, diarrhea, rash, skin and mucosal bleeding, headache, convulsion / seizure, and impaired consciousness) through the medical institution information management platform. Case information generally includes gender, date of birth, age, current address, occupation, time of onset of clinical symptoms, duration of symptoms, outpatient diagnosis, disease code, admission status, admission diagnosis, disease code, date of main diagnosis, admission route, admission time, discharge time, actual number of days in hospital, whether critically ill or seriously ill, admission condition, discharge condition, and hospitalization expenses.
[0021] Step 2. Clean the data extracted in step 1, assign values to relevant symptoms, count the number of visits for each symptom on a daily basis, and form a data set with the number of visits for the symptom as a feature.
[0022] Step 3. Input the data set in step 3 into the XGBoost model to predict the future number of cases of the target infectious disease and determine whether it will break out.
[0023] Examples:
[0024] This example is based on a case of a patient with symptoms related to a new respiratory infectious disease who visited a medical institution in a certain city.
[0025] 1. Data sources include: case information on relevant symptoms in medical institutions in a certain city in 2023 and report data on newly emerging respiratory infectious diseases in a certain city in 2023.
[0026] 2. Analysis process and explanation
[0027] 2.1 Extracting Data
[0028] Because of the privacy of the cases, this step was completed by an engineer from a city's disease control big data platform. The specific steps are that the engineer uses the symptom list provided by the researcher as a search term, extracts case information containing these symptoms in the main complaint, and performs desensitization. This example takes a new respiratory infectious disease as an example, so the symptom list is adjusted to sore throat, cough, myalgia, joint pain, back pain, headache, fever, dyspnea, fatigue, loss of smell, and loss of taste.
[0029] 2.2 Data cleaning, assignment, and statistics
[0030] (1) In theory, each person generates a unique visit number each time they visit the doctor, and the symptom monitoring information is unique. In reality, a variable may have multiple values, but all other variable information is the same. In this case, it is necessary to remove duplicates and only keep one piece of information. For example:
[0031]
[0032]
[0033] (2) Symptom identification and assignment
[0034] After data cleaning is completed, assign values to symptoms. Taking fever as an example, if the chief complaint includes "fever subsides|body temperature recovers|chills and fever|obvious fever|fever|fever|elevated body temperature|low-grade fever|high fever|abnormal body temperature|elevated body temperature|high body temperature|body fever|fever subsides", and does not include "no chills, fever|no chills and fever|no chills and fever|no chills and fever|no obvious fever|no obvious fever...", then the value is assigned to 1, otherwise it is assigned to 0. For example:
[0035]
[0036] (3) Statistics of daily visits for related symptoms
[0037] The number of visits for related symptoms on a daily basis was counted according to the date of visit. The statistical results are shown in Table 1.
[0038] Table 1 Daily visits to outpatient clinics for related symptoms in a certain city
[0039]
[0040]
[0041] 2.3 Prediction of infectious disease outbreak risk
[0042] (1) Model introduction
[0043] XGBoost, the full name of which is eXtreme Gradient Boosting, is an optimized distributed gradient boosting library designed to be efficient, flexible, and portable. It implements machine learning algorithms under the gradient boosting framework. XGBoost provides parallel tree boosting (also known as GBDT, GBM) to solve many data science problems quickly and accurately. The same code runs on major distributed environments (Hadoop, SGE, MPI) and can solve problems beyond billions of examples.
[0044] (2) Model training
[0045] The data in Table 1 are integrated with the report data of a new respiratory infectious disease to form a data set data (Table 2) for training the model. The first column is the date of consultation, the second to 13th columns are the number of consultations for each symptom (features), and the 14th column is the number of reported cases (labels).
[0046] Table 2 XGBoost model dataset composition
[0047]
[0048] (3) Model prediction effect
[0049] The root mean square error and mean absolute error are used as evaluation indicators. The smaller the value, the better the model fitting effect.
[0050] Table 3 Performance of XgBoost on training set and test set
[0051]
[0052] (4) Prediction results
[0053] The final output of this method is the number of infectious disease cases on the day corresponding to the symptom consultation date. According to relevant research on the outbreak of a new respiratory infectious disease, 30 confirmed cases mark the transition point from the slow growth stage to the rapid growth stage of a new respiratory infectious disease. Therefore, the early outbreak threshold of a new respiratory infectious disease is 30 cases.
[0054] in conclusion
[0055] The existing technology for predicting infectious diseases based on symptom monitoring is too broad and difficult to apply. This method uses mature medical institution information systems to extract information such as symptoms from electronic medical records, and uses machine learning methods to model and predict infectious disease outbreaks. Specific symptoms can be adjusted and combined according to the infectious diseases to be predicted, and the outbreak threshold can also be determined according to actual conditions.
[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for predicting the risk of infectious disease outbreaks based on electronic medical records. It is characterized in that The following steps are involved: Step 1: Extract case information containing symptoms related to the target infectious disease from the electronic medical records of past medical institutions to obtain the initial training data set; Step 2: Clean, assign values, and count the initial training data set to obtain the number of visits for each symptom on each visit date as a feature training data set, and collect the actual number of reported cases on each visit date as a label; Step 3: Input the feature training data set into the XGBoost model for training. The model outputs the predicted number of cases. By minimizing the error between the number of cases predicted by the model and the number of actual reported cases, the trained target infectious disease outbreak risk prediction XGBoost model is obtained. Step 4: Extract case information containing symptoms related to the target infectious disease from the chief complaint in the electronic medical records of medical institutions on that day to obtain the initial dataset to be predicted; perform data cleaning, value assignment, and statistics on the initial dataset to be predicted to obtain the number of visits for each symptom on that day as the feature dataset. Step 5: Input the feature data set into the trained target infectious disease outbreak risk prediction XGBoost model, output the case data prediction result of the day, and compare the prediction result with the preset threshold. If it is greater than the preset threshold, it is judged that the target infectious disease has broken out; otherwise, it is judged that the target infectious disease has not broken out.
2. The method for predicting the risk of infectious disease outbreaks based on electronic medical records according to claim 1, It is characterized in that In step 1, the symptoms related to the infectious disease are determined according to the symptoms that may be caused by the disease. A variety of symptom combinations are obtained by arranging and combining the symptoms. In step 2, feature training data sets of different symptom combinations are obtained respectively. In step 3, the XGBoost model training results corresponding to the feature training data sets of different symptom combinations are obtained, and the XGBoost model with the best effect is selected as the final training model.
3. The method for predicting the risk of infectious disease outbreaks based on electronic medical records according to claim 1, It is characterized in that The error in step 3 includes the root mean square error and the mean absolute error.
4. The method for predicting the risk of infectious disease outbreaks based on electronic medical records according to claim 1, It is characterized in that The step 2 specifically includes the following steps: Clean up duplicate medical information, assign 0 to the medical information containing the corresponding symptoms, otherwise assign 1.
5. The method for predicting infectious disease outbreak risk based on electronic medical records according to claim 1, It is characterized in that In step 2, each feature training data includes the date of consultation, the number of consultations corresponding to each symptom in the symptom combination, and the actual number of reported cases on that day.
