TWI751683B - Pathological condition prediction system for elderly flu patients, a program product thereof, and a method for establishing and using the same - Google Patents

Abstract

A pathological condition prediction system for elderly flu patients, a program product thereof, and a method for establishing and using the same. Establishing the pathological condition prediction system involves acquiring flu medical data form a medical database, screening out therefrom elderly flu patients who were 65 years old or above when arriving at hospitals and then weeding out the patients who had OHCA, and conducting AI learning to obtain prediction models; and loading a server host with a medical information service interface, a feature-capturing service program, and a prediction service program. In use, the feature-capturing service program extracts a medical feature value of an elderly flu patient from the medical database, and the prediction service program makes prediction using the prediction models according to the medical feature value, after which a prediction result is returned to the medical information service interface for a physician’s reference.

Description

Elderly influenza disease prediction system, program product and its establishment and use method

The present invention relates to an advanced influenza disease prediction system, a program product, and a method for establishing and using the same, in particular to an invention that uses AI to predict the development of the disease of an elderly influenza patient, so as to assist physicians in the follow-up treatment of the elderly influenza patient.

Taiwan has one of the fastest aging populations in the world. In 2018, Taiwanese elderly accounted for 14% of the total population, and it is expected to rapidly rise to 20% by 2025. From the clinical experience of the emergency department, it can be seen that the changes of the elderly’s condition are often complex and rapid, and it is also challenging to properly treat them in a timely manner. Influenza is a common seasonal disease in the elderly. In severe cases, hospitalization may be required, and sepsis may also be caused during the process. , secondary bacterial infection, or complications such as respiratory and ischemic heart disease, resulting in the death of elderly influenza patients.

Due to limited medical resources during the flu season, predicting the prognosis of elderly patients with flu and their subsequent management has become a very important topic.

In recent years, artificial intelligence (AI) has flourished, including technologies such as machine learning and natural language, which can process more variables without limiting the distribution of data. Therefore, the introduction of AI technology and the establishment of disease prediction models through electronic health records will enable Provide physicians with a better patient prognosis reference.

Chinese Patent No. CN110051324A proposes a "method and system for predicting mortality of acute respiratory distress syndrome", which is to predict mortality of acute respiratory distress syndrome through AI.

Based on the uncertainty of the disease development of the elderly influenza patients, the present invention finds out the influencing variables that affect the development of the elderly influenza patients, and uses the AI learning method to predict the development of the disease of the elderly influenza patients, so as to assist the physician in the Follow-up care for elderly influenza patients.

Therefore, the present invention proposes a method for establishing a system for predicting the condition of influenza in the elderly, comprising: The step of acquiring influenza medical data: extracting influenza medical data related to influenza from a raw data of a medical database.

The steps of AI learning for model training: Screen out the elderly influenza patients who are older than 65 years old in the above influenza medical data, and exclude the elderly influenza patients whose cardiopulmonary function stops before hospitalization; further clean the influenza medical data Convert to obtain complex characteristic variables into a large data database, and perform model training with AI according to the aforementioned characteristic variables, which include vital signs, medical history, patient action status and blood test values.

The step of obtaining the prediction model: According to the aforementioned model training, the prediction model is obtained, and the prediction model includes the probability of hospitalization, the probability of pneumonia, the probability of sepsis or shock, the probability of being transferred to an intensive care unit, and the probability of death.

Steps of establishing a network service: providing a medical information system service interface, a feature value extraction service program and a condition prediction service program; the medical information system service interface is linked to a medical information system for the medical information system to call the medical treatment an information system service interface, and make the feature value extraction service program retrieve a medical feature value of an elderly influenza patient from the medical database, the disease prediction The service program predicts with the above-mentioned prediction model according to the medical characteristic value, and returns a prediction result to the service interface of the medical information system.

Further, the cleaning and conversion of the influenza medical data is to modify the influenza medical data that does not conform to a standard data type to conform to the standard data type. Further, those that do not conform to the standard data type include incomplete data, mixed content, duplicate data, incorrect data generated without checking during input, incorrect format, null value, or one or a combination of different reporting units for different inspection methods .

Further, among the characteristic variables: vital signs include respiratory rate and coma index; medical history includes hypertension, coronary artery disease and malignant tumor; patient's action status is bedridden; blood test value includes white blood cell count, rod nucleus granulocyte, hemoglobin and C-reactive protein.

Further, the algorithm of the AI learning uses Random Forest (Random Forest), Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Multilayer Perceptron (Multilayer Perceptron) , MLP), Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (eXtreme Gradient Boosting, XGBoost), Logistic Regression (Logistic Regression). Further, in the AI learning process, the influenza medical data is divided into training set and test set, and 70% of the influenza medical data is used for the training set and 30% is used for the test set. This test set is validated.

Further, the prediction model also includes the probability of returning to the emergency department within one day, the probability of returning to the emergency department within three days, and the probability of returning to the emergency department within fourteen days.

Further, the prediction result is displayed side by side on the service interface of the medical information system in the form of visual graphs and figures.

The present invention further proposes a system for predicting the disease condition of the elderly by using the above-mentioned method for establishing the disease prediction system for the disease state of influenza in the elderly. and the medical information system are jointly connected to the medical database, the server host provides the medical information system service interface to the medical information system, and executes the feature value retrieval service program and the disease prediction service program.

Further, a physiological monitoring instrument is connected to the medical database.

The present invention further provides a program product, which is to load an application program into a computer to construct the above-mentioned system for predicting the condition of influenza in the elderly.

The present invention further provides a method for using the above-mentioned system for predicting the condition of influenza in the elderly, comprising: calling the medical information system service interface from the medical information system. The feature value retrieval service program retrieves the medical feature value of the elderly influenza patient from the medical database. The disease prediction service program predicts with the above-mentioned prediction model according to the medical characteristic value. The prediction result is returned to the service interface of the medical information system.

Further, the disease prediction service program uses a plurality of different algorithms to make predictions, and returns at most the same prediction results.

Further, a physiological monitoring instrument is used to obtain the medical characteristic value of the elderly influenza patient at any time, so that the disease prediction service program can continuously predict.

Through the above technical features, the following effects can be achieved:

1. The present invention finds out the influencing factors that affect the development of the elderly influenza disease, including respiratory rate, coma index, hypertension, coronary artery disease, malignant tumor, bed rest, white blood cell count, rod-shaped Nuclear granulocytes, heme, and C-reactive protein, and AI learning with the above-mentioned influencing factors as characteristic variables, so as to use AI to predict the development of the disease of elderly influenza patients, to assist physicians, especially emergency physicians in the treatment of elderly influenza patients. Follow-up treatment to improve the cure rate of elderly influenza patients.

2. In the AI learning process of the present invention, 70% of the medical data is used for the training set and 30% is used for the test set, and the test set is used for verification, and the accuracy rate (Accuracy), The sensitivity (Sensitivity), specificity (Specificity), and AUC are all high, and the physicians who apply it in the clinic are highly satisfied.

3. The present invention further obtains the physiological data related to the medical characteristic value of the elderly influenza patients at any time through the physiological monitoring instrument, so that the disease prediction service program can continuously predict, which is conducive to immediate correct treatment and reduces the occurrence of complications.

4. The present invention can use a plurality of different algorithms for prediction, and return at most the same prediction results to improve the accuracy of AI prediction.

1: Servo host

11: Big Data Repository

12: Medical Information System Service Interface

13: Feature value extraction service program

14: Condition prediction service program

2: Medical Information System

21: Medical record interface

22: Link Command

3: Medical database

31: External Information

32: In-hospital medical records

33: In-hospital structural medical data

34: In-hospital unstructured medical data

4: Physiological Monitoring Instruments

A: characteristic variable

A1: Respiratory rate

A2: Coma index

A3: Hypertension

A4: Coronary artery disease

A5: Malignant tumor

A6: Bed rest

A7: Count of white blood cells

A8: rod-shaped granulocyte

A9: Heme

A10:C-reactive protein

B: Algorithm

B1: Random Forest Algorithm

B2: Support Vector Machines

B3: K-proximity algorithm

B4: Multilayer Perceptron

B5: Lightweight Gradient Boosting Model

B6: Extreme Gradient Boosting

B7: Logistic regression analysis

C: Predictive model

C1: Rate of transfer to hospital

C2: probability of concurrent pneumonia

C3: Probability of complicated sepsis or shock

C4: Probability of transfer to ICU

C5: Chance of death

C6: Chance of returning to the emergency room again within a day

C7: Chance of returning to the emergency room again within three days

C8: Probability of returning to the emergency department within 14 days

[Figure 1] is a schematic diagram of the overall structure of the system for predicting the condition of influenza in the elderly according to an embodiment of the present invention.

[Fig. 2] is a schematic diagram of the establishment and prediction of a system for predicting the condition of influenza in the elderly according to an embodiment of the present invention.

[Figure 3] is a schematic diagram of a medical information system service interface in which a medical information system according to an embodiment of the present invention can be connected to a server host.

[FIG. 4] is a schematic diagram of a service interface of a medical information system of a server host according to an embodiment of the present invention.

[FIG. 5] is a schematic diagram of continuously monitoring the physiological data related to medical characteristic values of elderly influenza patients with a physiological monitoring instrument for continuous prediction in an embodiment of the present invention.

In view of the above technical features, the main effects of the system for predicting the condition of influenza in the elderly, the program product and the method for establishing and using the same of the present invention will be clearly presented in the following examples.

Referring to the first figure, the system for predicting the condition of flu in the elderly includes a server host 1 for AI prediction. The server host 1 has a large data database 11 and can provide a medical information system service interface 12, a A feature value extraction service program 13 and a disease prediction service program 14; the server host 1 is connected to a medical information system 2 (HIS) of a medical institution, and the server host 1 and the medical information system 2 are jointly connected to a medical database 3.

Referring to Figures 1 and 2, the establishment of the above-mentioned system for predicting the condition of influenza in the elderly requires the establishment of the server host 1 first, which specifically includes the following steps: the steps of acquiring influenza medical data-

Influenza medical data related to influenza is retrieved from a raw data of the medical database 3 . The medical database 3 may include: health insurance database, death data of the National Construction Agency, external data such as health screening tests 31 , in-hospital medical record data 32 of individual hospitals, in-hospital structural medical data 33 of individual hospitals, and imaging studies of individual hospitals Data, photos, texts and other in-hospital non-structural medical data34, etc., the influenza medical data in this example comes from the data from the General Hospital of Chi Mei Hospital, Liuying Hospital District, and Jiali Hospital District from 2009 to 2018, and Obtained from the emergency medical order, nursing record, appraisal, medical history, and inspection system established by the medical information system 2 in the hospital.

The steps of AI learning for model training-

According to the definition of the United Nations World Health Organization (WHO), the population over 65 years old is the elderly population. Therefore, in the above influenza medical data, we screen out the elderly influenza patients who are older than 65 years old, and exclude those who have stopped cardiopulmonary function before hospitalization. Influenza patients. In emergency patients with influenza, the doctor will first prescribe influenza drugs and enter the relevant diagnosis of influenza in the diagnosis field, but sometimes in emergency situations, the doctor will first deal with the relevant treatment verbally, so it is necessary to determine whether it is elderly influenza Patients can be extracted from a doctor's order diagnosis or a flu drug pricing system.

Further, multiple characteristic variables A are obtained from the influenza medical data and stored in the big data database 11. According to the emergency physicians of Chi Mei Hospital, about 6,000 elderly influenza patients were recruited in the past ten years, and AI was used for model training with the aforementioned characteristic variables A. . The selection of the above characteristic variables A is based on the analysis of ten factors that may affect the development of the disease. These characteristic variables A can be classified into vital signs, medical history, patient action status and blood test values. Among them: vital signs include respiratory rate A1 (Tachypnea ,respiratory rate) and coma index A2 (Severe coma, Glasgow Coma Scale); medical history includes hypertension A3 (Hypertension), coronary artery disease A4 and malignant tumor A5 (Cancer); the patient's action status is bedridden A6 ( Bedridden); blood test values include A7 (Leukocytosis, white blood cell count), A8 (Bandemia, white blood cell count band form), heme A9 (Anemia, hemoglobin) and C-reactive protein A10 (Elevated CRP, C-reactive protein). Among them, the respiratory rate A1 is more than 20 times per minute; the coma index A2 is less than 8; the white blood cell count A7 is more than 12,000 per CC; the rod-like granulocyte A8 is more than 10 percent; less than 12mg/dL; C-reactive protein A10 more than 10mg/dL.

When the above-mentioned influenza medical data does not conform to a standard data type, the influenza medical data is first cleaned and converted to conform to the standard data type. Those that do not conform to the standard data type, such as incomplete data, mixed content, duplicate data, incorrect data generated without checking during input, format incorrect formula, null values, or different reporting units for different test methods, etc. The cleaning and conversion of the influenza medical data may include giving the blank part of the incomplete data to the value that should be filled in according to clinical practical experience, for example, the respiratory rate A1 is represented by 12; the coma index A2 is represented by 15; the white blood cell count A7 is represented by The number of each CC is expressed as 7000; the rod-shaped granulocyte A8 is expressed as 0%; the heme A9 is expressed as 12g/dL; the C-reactive protein A10 is expressed as 2.5mg/L.

The above-mentioned characteristic variable A is learned by AI for statistical classification, and the algorithm B used in the AI learning in this embodiment includes random forest algorithm B1 (Random Forest), support vector machine B2 (Support Vector Machines, SVM), K - Proximity Algorithm B3 (K Nearest Neighbor, KNN), Multilayer Perceptron B4 (Multilayer Perceptron, MLP), Light Gradient Boosting Model B5 (Light Gradient Boosting Machine, LightGBM), Extreme Gradient Boosting B6 (eXtreme Gradient Boosting, XGBoost) ), logistic regression analysis B7 (Logistic Regression). In the AI learning process, the influenza medical data is divided into a training set and a test set, and 70% of the influenza medical data is used for the training set and 30% is used for the test set, and the test set is used. authenticating. Refer to Table 1 to Table 8 below. According to the above AI learning and verification results, its prediction accuracy (Accuracy), sensitivity (Sensitivity), specificity (Specificity), and AUC (area under the curve) are as high as 70% to 90% %above.

Table 6: Probability of Complicated Sepsis or Shock

Steps to obtain a predictive model-

According to the aforementioned model training, a subsequent prediction model C for AI prediction is obtained. The prediction model C includes the probability of hospitalization transfer C1, the probability of pneumonia complicated by C2, the probability of complicated sepsis or shock C3, the probability of transfer to the intensive care unit C4, the probability of death C5, and the probability of death within one day. There are eight prediction models including the probability of returning to the emergency department again C6, the probability of returning to the emergency department again within three days C7 and the probability of returning to the emergency department again within 14 days C8.

Steps to create a web service-

After the above-mentioned prediction model C is obtained, the medical information system service interface 12 , the feature value extraction service program 13 and the disease prediction service program 14 required for the above-mentioned AI prediction can be established into a program product, through which the program product will The application program is loaded into a computer as the server host 1 . After the server host 1 is set up, the server host 1 can be connected to the medical information system 2 (HIS) of the medical institution and the aforementioned medical database 3 .

Referring to Figures 3 and 4, when a patient sees a doctor, the doctor will input relevant medical data in a medical record interface 21 of the medical information system 2 of the medical institution where he is located, and these medical data will be stored in the relevant medical data In the library 3, there is a link command 22 for predicting the condition of influenza in the elderly in the medical record interface 21. When the doctor determines that the patient to be seen is an elderly flu patient, the doctor can click on the link command 22, and the medical information system 2 will call the service interface 12 of the medical information system. At this time, the medical information system will be displayed on the doctor's consultation computer. The information system service interface 12 executes the characteristic value extraction service program 13 and the disease prediction service program 14, the characteristic value extraction service program 13 extracts the medical treatment of the elderly influenza patient from the relevant medical database 3 In the data, the medical characteristic value related to the aforementioned characteristic variable A is then predicted by the disease prediction service program 14 according to the medical characteristic value using the above-mentioned prediction model, and the prediction result is returned to the medical information system service interface 12, and The prediction result can be displayed on the service interface 12 of the medical information system in the form of visual graphics and numbers to facilitate observation, and the prediction result can help doctors to understand the possible disease development of the elderly influenza patient, so as to facilitate follow-up treatment , to improve the cure rate of elderly influenza patients.

Among them, the disease prediction service program 14 can choose to use random forest algorithm B1 (Random Forest), support vector machine B2 (Support Vector Machines, SVM), K-nearest algorithm B3 (K Nearest Neighbor, KNN), multi-layer perceptron B4 (Multilayer Perceptron, MLP), Lightweight Gradient Boosting Model B5 (Light Gradient Boosting Machine, LightGBM), Extreme Gradient Boosting One of B6 (eXtreme Gradient Boosting, XGBoost) and logistic regression analysis B7 (Logistic Regression) is used for prediction. In order to improve the prediction accuracy, all algorithms can be used for prediction at the same time, and at most the same prediction results are returned.

In the embodiment of the present invention, a total of 84 historical records of influenza in the elderly were tested by the General Hospital of Chi Mei Hospital from January to March 2019. The test results were filled in by physicians for satisfaction. A total of 51 records were recovered and scored on a 5-point scale. Satisfaction is 4.6 points, showing the feasibility of the system.

Referring to Figures 1 and 5, a physiological monitoring instrument 4, such as a sphygmomanometer, oximeter, etc., can be used to obtain the medical data of the elderly influenza patient at any time during the diagnosis and treatment of the elderly influenza patient, and upload them to the relevant In the medical database 3, the feature value retrieval service program 13 will continuously or periodically retrieve the relevant medical feature values from the medical database 3, and then the disease prediction service program 14 will continuously perform prediction, which is conducive to real-time implementation. With the correct treatment, reduce the occurrence of complications.

Based on the descriptions of the above embodiments, one can fully understand the operation, use and effects of the present invention, but the above-mentioned embodiments are only preferred embodiments of the present invention, which should not limit the implementation of the present invention. Scope, that is, simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the description of the invention, all fall within the scope of the present invention.

1: Servo host

12: Medical Information System Service Interface

13: Feature value extraction service program

14: Condition prediction service program

2: Medical Information System

3: Medical database

4: Physiological Monitoring Instruments

C: Predictive model

Claims (13)

A method for establishing a system for predicting influenza disease conditions in the elderly, comprising: the step of acquiring influenza medical data: extracting influenza medical data related to influenza from a raw data of a medical database; the step of AI learning and model training: in the above-mentioned steps From the influenza medical data, screen out the elderly influenza patients who are older than 65 years old, and exclude the elderly influenza patients with pre-hospital cardiopulmonary function arrest; further clean and convert the influenza medical data to obtain the complex characteristic variables into a big data Database, model training is performed with AI according to the aforementioned characteristic variables. The characteristic variables include vital signs, medical history, patient action status and blood test values. Among the characteristic variables: vital signs include respiratory rate and coma index; medical history includes hypertension , coronary artery disease and malignant tumor; the patient's action status is bedridden; the blood test value includes white blood cell count, rod-shaped granulocyte, heme and C-reactive protein; the steps of obtaining the prediction model: the prediction model is obtained according to the aforementioned model training. The prediction model includes the probability of being transferred to hospital, the probability of pneumonia, the probability of complicated sepsis or shock, the probability of being transferred to the intensive care unit, and the probability of death; the steps of establishing a network service: providing a medical information system service interface, a feature value extraction service program and a Condition prediction service program; the medical information system service interface is linked to a medical information system for the medical information system to call the medical information system service interface, and the feature value retrieval service program retrieves an elderly influenza from the medical database A medical characteristic value of the patient, the disease prediction service program predicts with the above-mentioned prediction model according to the medical characteristic value, and returns a prediction result to the service interface of the medical information system. As claimed in claim 1, the method for establishing an influenza disease prediction system for the elderly, wherein the cleaning and conversion of the influenza medical data is to modify the influenza medical data that does not conform to a standard data type to conform to the standard data type. For example, the method for establishing an influenza disease prediction system for the elderly in claim 2, the data types that do not conform to the standard include incomplete data, mixed content, duplicate data, wrong data generated without checking when input, incorrect format, empty data, etc. One or a combination of different reporting units for values or different test methods. The method for establishing a system for predicting the condition of influenza in the elderly according to claim 1, wherein the AI learning algorithm is a random forest algorithm (Random Forest), Support Vector Machines (SVM), K-neighbor algorithm (K Nearest Neighbor, KNN), Multilayer Perceptron (MLP), Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (eXtreme Gradient Boosting, XGBoost), Logistic Regression (Logistic Regression) )one. As in claim 4, the method for establishing an influenza disease prediction system for the elderly, wherein, in the AI learning process, the influenza medical data is divided into a training set and a test set, and 70% of the influenza medical data is used for the training set , 30% is used for the test set, and the test set is used for validation. As claimed in claim 1, the method for establishing an influenza disease prediction system for the elderly, wherein the prediction model further includes the probability of returning to the emergency department within one day, the probability of returning to the emergency department within three days, and the probability of returning to the emergency department within fourteen days. As claimed in claim 1, the method for establishing an influenza disease prediction system for the elderly, wherein the prediction result is displayed side by side in the service interface of the medical information system in the form of visual graphs and numbers. A system for predicting influenza disease condition in the elderly established by using the method for establishing a system for predicting influenza disease condition in the elderly according to any one of claim 1 to claim 7, comprising: a server host having the big data database, and the server host connecting to the medical information system, and the server host and the medical information system are jointly connected to the medical database, the server host provides the medical information system service interface to the medical information system, and executes the feature value extraction service program and the disease prediction service program. If the system for predicting the condition of influenza in the elderly according to claim 8, further, a physiological monitoring instrument is connected to the medical database. A program product, which is to load an application program into a computer, so as to construct the system for predicting the condition of influenza in the elderly as claimed in item 8. A method for using a system for predicting the condition of an elderly flu disease according to claim 8, comprising: calling a service interface of the medical information system from the medical information system; the medical characteristic value; the disease prediction service program predicts with the above-mentioned prediction model according to the medical characteristic value; and returns the prediction result to the service interface of the medical information system. As claimed in claim 11, for the method of using a system for predicting an influenza disease condition in the elderly, wherein the disease prediction service program uses a plurality of different algorithms to make predictions, and returns at most the same prediction results. According to the method for using the system for predicting the condition of flu in the elderly according to claim 11, further, a physiological monitoring instrument is used to obtain the medical characteristic value of the flu patient of the elderly at any time, so that the condition prediction service program can continuously make predictions.

Images