Influenza prevalence trend prediction and early warning method

Abstract

The invention provides an influenza prevalence trend prediction and early warning method, which comprises the following steps of S1, collecting multi-source data consisting of monitored epidemic situation data, monitored event data and meteorological data, analyzing influenza epidemiological characteristics, and screening influenza prediction and early warning indexes; s2, analyzing the influenza epidemic situation, and predicting the influenza epidemic trend; and S3, analyzing the flu prevalence signal, and performing early warning on the flu prevalence abnormal signal. The method has the beneficial effects that the prediction and early warning model of the influenza epidemiology trend is established by analyzing the influenza epidemiology characteristics, and reference is provided for the public and related health departments to deal with the influenza in time.

Description

Technical Field

[0001] The invention belongs to the field of disease prediction and early warning, and in particular relates to a method for predicting and early warning of influenza epidemic trends. Background Art

[0002] Influenza, also known as flu, is an acute respiratory infectious disease caused by influenza virus. The main clinical manifestations of influenza are poisoning symptoms such as high fever, fatigue, headache, and body muscle aches, while respiratory symptoms are mild. The incubation period of influenza is short, ranging from 1 to 3 days, with the shortest being a few hours and the longest being up to 4 days. It is mainly transmitted through the respiratory tract, is highly contagious, and spreads quickly. Influenza viruses are prone to mutation, resulting in widespread susceptibility to the disease and a high incidence rate.

[0003] China is a country with a high burden of influenza infection. A population-based study of deaths from influenza virus respiratory infections showed that an average of 88,000 influenza deaths occur in my country each year, accounting for about 13.6% of the global influenza deaths

[11] . Current influenza prevention and control still faces severe challenges. Research on influenza prediction and early warning is of great significance in the field of public health and disease control.

[0004] The current limitations of influenza prediction and early warning research are manifested in two aspects: (1) Incomplete data use: Currently, there are only a few studies that can use multiple types of data for influenza prediction at the same time. The possible reasons are, on the one hand, the difficulty of data acquisition, and on the other hand, the current prediction and early warning models cannot handle multi-source heterogeneous data. (2) Low model performance accuracy: A single model has poor generalization and robustness and can only perform well in one aspect; the thresholds for influenza epidemics are different in different seasons and regions, with a high degree of variation, and the thresholds are difficult to determine. Summary of the invention

[0005] In view of this, the present invention aims to provide a method for predicting and warning influenza epidemic trends.

[0006] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0007] A method for predicting and warning influenza epidemic trends, comprising the following steps:

[0008] S1, collect multi-source data consisting of epidemic monitoring data, event monitoring data and meteorological data, analyze the epidemiological characteristics of influenza, screen indicators for influenza prediction and early warning, and obtain independent variables;

[0009] S2, analyze the epidemic situation of influenza and predict the epidemic trend of influenza;

[0010] S3, analyze the signals of influenza epidemic and issue early warning for abnormal signals of influenza epidemic.

[0011] Furthermore, the multi-source data in S1 includes:

[0012] (1) Epidemic surveillance data: percentage of influenza-like illness in the north and south, influenza virus positivity rate in the north and south; percentage of influenza-like illness in the region, influenza virus positivity rate in the region; regional influenza incidence data;

[0013] (2) Monitoring event data: regional influenza-related drug sales, regional school absenteeism data; Internet keyword search volume data;

[0014] (3) Meteorological data: regional daily average temperature, average humidity, average precipitation, and average air pressure.

[0015] Furthermore, when analyzing the epidemiological characteristics of influenza, we used the monitoring epidemic data from multi-source data to analyze the time, scale and peak of influenza epidemics through descriptive analysis and seasonal decomposition methods;

[0016] Through descriptive analysis, the average, maximum and standard deviation of the data in the region are obtained, and the trend of influenza is visually displayed through a time series graph;

[0017] Through seasonal decomposition, long-term trends and seasonal patterns of influenza are derived.

[0018] Furthermore, the monitoring event data and meteorological data are analyzed to screen indicators for influenza prediction and early warning;

[0019] The correlation analysis function in the social science statistical software SPSS was used to calculate the correlation coefficient between the monitoring event data, meteorological data and the percentage of regional influenza-like illness. If the correlation coefficient was > 0.4, it was initially selected for inclusion in the model.

[0020] The data obtained through cross-correlation analysis were then used to perform the lasso algorithm using R software to screen out independent variables suitable for predicting influenza trends;

[0021] Finally, the screened independent variables are used as indicators for influenza prediction and early warning.

[0022] Furthermore, the Internet keywords include: gastrointestinal cold, what is influenza A, bronchitis, influenza, sneezing, influenza vaccine, and 999 cold medicine.

[0023] Furthermore, in S2, the data screened in S1 are used as independent variables, and the regional influenza-like illness percentage is used as the dependent variable to construct a prediction model for the regional influenza epidemic trend, epidemic time and epidemic peak.

[0024] Further, the prediction model training method:

[0025] (1) Data preparation: Organize the collected multi-source data in time series.

[0026] (2) Data division: Divide the data into training set, validation set and test set.

[0027] (3) Model design and training: Build the LSTM model architecture, set hyperparameters such as the number of hidden layers and nodes, use the training set data to train the model, and adjust the model parameters through the validation set.

[0028] (4) Prediction: Use the trained LSTM model to predict influenza trends, epidemic time, and epidemic peak.

[0029] Furthermore, in S3, based on the prediction results of S2, the percentage of regional influenza-like illness is used as the influenza indicator to calculate the threshold of influenza epidemic, capture abnormal signals, and complete the early warning of abnormal signals of influenza epidemic.

[0030] Furthermore, the moving percentile method is used to calculate the threshold of influenza epidemic:

[0031] (1) Calculate the threshold: divide the historical data into several windows according to time, and calculate the percentile of the influenza indicator in each window;

[0032] (2) Set the warning threshold: Select an appropriate percentile as the warning threshold;

[0033] (3) Capturing abnormal signals: Compare the current influenza indicators with the real-time monitoring data. If they exceed the warning threshold, it is considered that an abnormal signal has been captured;

[0034] (4) Complete early warning: Issue influenza epidemic warnings based on abnormal signals and combined with forecast results.

[0035] The advantages and positive effects of the present invention are as follows: the present invention establishes a prediction and early warning model for influenza epidemic trends by analyzing the epidemiological characteristics of influenza, thereby providing a reference for the public and relevant health departments to respond to influenza in a timely manner. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the 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. In the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0037] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0038] The present invention is a method for predicting and warning influenza epidemic trends, comprising the following steps:

[0039] S1, collect multi-source data consisting of epidemic monitoring data, event monitoring data and meteorological data, analyze the epidemiological characteristics of influenza, screen indicators for influenza prediction and early warning, and obtain independent variables;

[0040] We collected data from multiple sources from January 1, 2012 to October 31, 2023, including:

[0041] (1) Epidemic surveillance data: data on the percentage of influenza-like illness and influenza virus positivity in the north and south obtained from the China National Influenza Center; data on the percentage of influenza-like illness and influenza virus positivity in Tianjin obtained from the Tianjin Center for Disease Control and Prevention; and data on the influenza incidence in Tianjin obtained from the Public Health Science Data Center of the Chinese Center for Disease Control and Prevention.

[0042] (2) Monitoring event data: data on sales of influenza-related drugs and number of absenteeism in schools in the Tianjin area obtained from the Tianjin Center for Disease Control and Prevention; data on Internet keyword search volume in the Tianjin area obtained from websites such as Webmaster Home and Baidu Index; Internet keywords include: stomach cold, what is influenza A, tracheitis, influenza, sneezing, influenza vaccine, 999 cold medicine, and other influenza-related Internet search terms.

[0043] (3) Meteorological data: Meteorological monitoring data such as daily average temperature, average humidity, average precipitation, and average air pressure in Tianjin obtained from the National Meteorological Science Data Center.

[0044] The multi-source data is shown in Table 1:

[0045]

[0046] Table 1

[0047] When analyzing the epidemiological characteristics of influenza, we used monitoring epidemic data from multiple sources to analyze the time, scale and peak of influenza epidemics through descriptive analysis and seasonal decomposition methods.

[0048] Through descriptive analysis, the average, maximum and standard deviation of the data in the region are obtained, and the trend of influenza is visually displayed through time series graphs, while seasonal decomposition further reveals the long-term trend and seasonal pattern of influenza.

[0049] Analysis results: Influenza usually peaks in winter, but there are certain differences in the peak time of influenza in the north and south. The peak in the south is often slightly later than that in the north, and there is only one winter influenza peak in the north. These findings are of great reference value for the formulation of vaccination strategies and public health policies.

[0050] Next, we analyzed the monitoring event data and meteorological data to screen indicators for influenza prediction and warning;

[0051] The correlation analysis function in the social science statistical software SPSS was used to calculate the correlation coefficient between the monitoring event data, meteorological data and the percentage of regional influenza-like illness. If the correlation coefficient was > 0.4, it was initially selected for inclusion in the model.

[0052] The data obtained through cross-correlation analysis were then used to execute the lasso algorithm using R software (RStudio software) to screen out independent variables suitable for predicting influenza trends; for example, if 13 variables were input, only 5 variables would remain after the lasso screening algorithm.

[0053] Finally, the screened independent variables are used as indicators for influenza prediction and early warning.

[0054] S2, analyze the epidemic situation of influenza and predict the epidemic trend of influenza;

[0055] Using the data screened in S1 as the independent variable and the regional influenza-like illness percentage as the dependent variable, a prediction model for influenza epidemic trend, epidemic time and epidemic peak was constructed.

[0056] Prediction model training method:

[0057] (1) Data preparation: Organize the collected multi-source data in time series.

[0058] (2) Data division: Divide the data into training set, validation set and test set.

[0059] (3) Model design and training: Build the LSTM model architecture. The long short-term memory network (LSTM) is a recurrent neural network (RNN) specifically designed for processing and predicting time series data. It can capture long-term and short-term dependencies in the sequence and is very suitable for processing the nonlinear trend of influenza data. Set hyperparameters such as the number of hidden layers and nodes, use the training set data for model training, and adjust the model parameters through the validation set.

[0060] The basic architecture of the LSTM model and its components:

[0061] a. Cell State: The core component of LSTM, used to transmit information throughout the chain. The cell state is similar to a conveyor belt, which can pass through the entire network relatively unchanged.

[0062] b. Gating mechanism (Gates): LSTM contains three main gating mechanisms to regulate the flow of information:

[0063] Forget Gate: Determines how much information from the cell state needs to be discarded. This is achieved through a sigmoid activation function and an element-wise product operation.

[0064] Input Gate: Determines how much input information will be stored in the cell state. Includes a sigmoid function to determine which information needs to be updated, and a tanh function to generate candidate update vectors.

[0065] Output Gate: Determines which parts of the cell state will be output to the next unit. Combines the sigmoid function and the tanh function to filter the cell state.

[0066] c. Cell state update: Through the above gates, LSTM can update the new hidden state and new cell state based on the current input, previous hidden state and cell state.

[0067] (4) Prediction: Use the trained LSTM model to predict influenza trends, epidemic time, and epidemic peak.

[0068] S3, analyze the signals of influenza epidemic and issue early warning for abnormal signals of influenza epidemic.

[0069] According to the prediction results of S2, the percentage of regional influenza-like illness is used as the research variable to calculate the threshold of influenza epidemic, capture abnormal signals, and complete the early warning of abnormal signals of influenza epidemic.

[0070] We use the moving percentile method to calculate the thresholds for influenza epidemics. The moving percentile method is a statistical method based on historical data to determine the thresholds for influenza epidemics. When the observed influenza indicators exceed these thresholds, an early warning signal is issued.

[0071] (1) Calculating the threshold: Divide the historical data into several windows according to time, and calculate the percentile (such as 90%, 95%, etc.) of the influenza indicator (such as the percentage of regional influenza-like illness) in each window (such as each week).

[0072] (2) Set the warning threshold: Select an appropriate percentile as the warning threshold.

[0073] (3) Capturing abnormal signals: Compare the current influenza indicators with the real-time monitoring data. If they exceed the warning threshold, it is considered that an abnormal signal has been captured.

[0074] (4) Complete early warning: Issue influenza epidemic warnings based on abnormal signals and combined with forecast results.

[0075] By combing through the literature related to influenza prediction and early warning, we discovered the limitations of previous research content (including the models used, indicators and reported results); we collected and organized multi-source data on influenza, analyzed the epidemiological characteristics of influenza, and screened indicators for influenza prediction and early warning; we explored the situation of influenza epidemics and predicted the epidemic trends of influenza; we analyzed the signals of influenza epidemics and issued early warnings for abnormal signals of influenza epidemics, which can provide a reference for the public and relevant health departments to respond to influenza epidemics in a timely manner.

[0076] Influenza not only has a high prevalence rate, but also a large number of severe cases, and the respiratory complications caused by influenza cannot be ignored. The situation of influenza prevention and control is severe, so it is necessary to grasp the trend of influenza epidemic in time and to predict and warn against influenza. The present invention is based on the current situation of the country, conforms to national policies and guidelines, and uses objective data for analysis, which has very important practical significance for studying the trend of influenza epidemic in China (Tianjin, etc.).

[0077] The embodiments of the present invention are described in detail above, but the contents described are only preferred embodiments of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for predicting and warning influenza epidemic trends, characterized by: The following steps are involved: S1, collect multi-source data consisting of epidemic monitoring data, event monitoring data and meteorological data, analyze the epidemiological characteristics of influenza, screen indicators for influenza prediction and early warning, and obtain independent variables; S2, analyze the epidemic situation of influenza and predict the epidemic trend of influenza; S3, analyze the signals of influenza epidemic and issue early warning for abnormal signals of influenza epidemic.

2. The influenza epidemic trend prediction and early warning method according to claim 1, characterized in that: The multi-source data in S1 includes: (1) Epidemic surveillance data: percentage of influenza-like illness in the north and south, influenza virus positivity rate in the north and south; percentage of influenza-like illness in the region, influenza virus positivity rate in the region; regional influenza incidence data; (2) Monitoring event data: regional influenza-related drug sales, regional school absenteeism data; Internet keyword search volume data; (3) Meteorological data: regional daily average temperature, average humidity, average precipitation, and average air pressure.

3. The influenza epidemic trend prediction and early warning method according to claim 2 is characterized by: When analyzing the epidemiological characteristics of influenza, we used monitoring epidemic data from multiple sources to analyze the time, scale and peak of influenza epidemics through descriptive analysis and seasonal decomposition methods. Through descriptive analysis, the average, maximum and standard deviation of the data in the region are obtained, and the trend of influenza is visually displayed through a time series graph; Through seasonal decomposition, long-term trends and seasonal patterns of influenza are derived.

4. The influenza epidemic trend prediction and early warning method according to claim 2, characterized in that: Analyze monitoring event data and meteorological data to screen indicators for influenza prediction and early warning; The correlation analysis function in the social science statistical software SPSS was used to calculate the correlation coefficient between the monitoring event data, meteorological data and the percentage of regional influenza-like illness. If the correlation coefficient was > 0.4, it was initially selected for inclusion in the model. The data obtained through cross-correlation analysis were then used to perform the lasso algorithm using R software to screen out independent variables suitable for predicting influenza trends; Finally, the screened independent variables are used as indicators for influenza prediction and early warning.

5. The influenza epidemic trend prediction and early warning method according to claim 2 is characterized by: The Internet keywords include: gastrointestinal cold, what is influenza A, tracheitis, influenza, sneezing, influenza vaccine, and 999 cold medicine.

6. The influenza epidemic trend prediction and early warning method according to claim 1, characterized in that: In S2, the data screened in S1 were used as independent variables, and the regional influenza-like illness percentage was used as the dependent variable to construct a prediction model for regional influenza epidemic trends, epidemic time, and epidemic peak.

7. The influenza epidemic trend prediction and early warning method according to claim 6, characterized in that: Prediction model training method: (1) Data preparation: Organize the collected multi-source data into time series; (2) Data partitioning: Divide the data into training set, validation set, and test set; (3) Model design and training: Build the LSTM model architecture, set hyperparameters such as the number of hidden layers and nodes, use the training set data to train the model, and adjust the model parameters through the validation set; (4) Prediction: Use the trained LSTM model to predict influenza trends, epidemic time, and epidemic peak.

8. The influenza epidemic trend prediction and early warning method according to claim 1, characterized in that: In S3, based on the prediction results of S2, the percentage of regional influenza-like illness is used as the influenza indicator to calculate the threshold of influenza epidemic, capture abnormal signals, and complete the early warning of abnormal signals of influenza epidemic.

9. The influenza epidemic trend prediction and early warning method according to claim 8, characterized in that: Calculate the influenza epidemic threshold using the moving percentile method: (1) Calculate the threshold: divide the historical data into several windows according to time, and calculate the percentile of the influenza indicator in each window; (2) Set the warning threshold: Select an appropriate percentile as the warning threshold; (3) Capturing abnormal signals: Compare the current influenza indicators with the real-time monitoring data. If they exceed the warning threshold, it is considered that an abnormal signal has been captured; (4) Complete early warning: Issue influenza epidemic warnings based on abnormal signals and combined with forecast results.

Images