A geographic weighting-based infectious disease risk analysis method and system

Abstract

This invention provides a geographically weighted infectious disease risk analysis method and system, comprising: rasterizing multi-source data and constructing a unified grid dataset and analysis indicator library; calculating the Pearson correlation coefficient between grid cases and various indicators in the unified grid dataset, and outputting the final set of influencing indicators; determining the neighborhood range using an adaptive bandwidth method and determining the spatial weights corresponding to neighborhood samples based on a specified kernel function; constructing exposure sequences corresponding to the current analysis period and multiple historical lag periods for each key indicator; forming a cross-basis local sample set based on the cross-basis features constructed based on the target grid cell and neighboring grid cells, constructing a geographically weighted distributed lag nonlinear model based on the cross-basis local sample set, and performing local regression fitting; comparing the effect differences of different exposure levels over all lag periods to obtain the cumulative effect value, and determining the lag period that contributes the most to the infectious disease risk of the target grid cell as the optimal lag period.

Description

Technical Field

[0001] This invention relates to the field of public health data processing technology, and in particular to a geographically weighted method and system for infectious disease risk analysis. Background Technology

[0002] With global climate change, accelerated urbanization, and more complex population movements, the occurrence and spread of infectious diseases exhibit increasingly complex spatiotemporal heterogeneity. Numerous studies have shown that factors such as temperature, precipitation, humidity, air pollutants, land cover, population density, population mobility, and social development levels can significantly impact the risk of infectious disease outbreaks by influencing pathogen survival, vector reproduction, population contact patterns, and accessibility to health resources. Furthermore, these impacts are often not immediate but rather exhibit significant lags and nonlinear characteristics. Therefore, in infectious disease risk assessment and early warning, simultaneously characterizing the influence of multiple factors, spatial differences, and lag effects has become an important research direction in this field. Currently, common methods for infectious disease risk analysis still have the following limitations: (1) Insufficient ability to process multi-source data uniformly: Existing infectious disease risk analysis methods usually focus on the statistical analysis of single-type data or a small number of influencing factors, making it difficult to integrate population data, infectious disease case data, environmental data, population mobility data and regional multidimensional indicator data under a unified framework. Due to the significant differences in spatiotemporal resolution and statistical caliber among different data sources, existing technologies often fail to achieve unified gridded processing and collaborative analysis, resulting in inconsistent analysis results at spatial scales and a lack of comparability between different types of influencing factors, thus restricting the level of refinement in infectious disease risk analysis.

[0003] (2) Insufficient spatial precision: Many existing methods for infectious disease risk research still use administrative regions as statistical units or employ overall regional models to analyze the relationship between cases and influencing factors. While these methods are relatively simple to implement, they assume that the exposure-onset relationship is the same across different spatial locations within the study area, making it difficult to reflect the local risk mechanisms formed between smaller spatial units due to differences in natural conditions, population density, mobility patterns, and socioeconomic background. For the pathogenesis of infectious diseases with significant spatial heterogeneity, using a uniform parameter model can easily obscure local high-risk areas and their dominant influencing factors, making it difficult to meet the needs of refined analysis and regional prevention and control.

[0004] (3) Insufficient screening of key influencing indicators: In existing technologies, infectious disease risk research mainly relies on empirical judgment or simple statistical screening when selecting variables, lacking a systematic screening process for the strength of correlations among multiple candidate indicators, and especially lacking technical means to connect correlation analysis with the subsequent modeling process. For multi-source variables such as environmental indicators, population mobility indicators, and socio-economic development indicators, if the correlation between them and gridded cases cannot be quantitatively evaluated first, and indicators with strong collinearity cannot be further eliminated, it is easy to lead to variable redundancy, unstable parameters, and insufficient interpretability of results in the model, thereby affecting the accuracy and reliability of risk analysis.

[0005] (4) Insufficient representation of lag nonlinear effects: The impact of environmental factors on the incidence of infectious diseases is usually not immediate, but gradually manifests over a certain period of time, and this relationship often has nonlinear characteristics. Although some existing methods can analyze the statistical association between environmental factors and cases, they often only consider the current period's impact or a single fixed lag, making it difficult to systematically express the continuous effect changes over multiple lag periods; other methods, although introducing distributed lag nonlinear models, are mostly based on the overall regional scale and still fail to take into account local differences in different spatial locations. Therefore, existing technologies generally cannot simultaneously characterize the three types of features: "spatial heterogeneity + nonlinear effect + lag effect," resulting in an insufficient description of the risk formation mechanism of infectious diseases.

[0006] Therefore, there is an urgent need to propose a technical solution for infectious disease risk analysis that can integrate multi-source data and complete the construction of analytical indicators, selection of key indicators, spatial weighted modeling, and lag nonlinear analysis within a unified grid framework. This solution aims to overcome the shortcomings of existing technologies in unified processing of multi-source data, expression of spatial heterogeneity, selection of key indicators, and characterization of lag nonlinearity, thereby enabling more refined analysis and assessment of infectious disease risks. Summary of the Invention

[0007] This invention provides a geographically weighted infectious disease risk analysis method and system to address the shortcomings of existing technologies, enabling the analysis and assessment of spatial heterogeneity and lag effects of infectious disease incidence risks, and providing technical support for infectious disease risk early warning and prevention.

[0008] In a first aspect, the present invention provides a geographically weighted infectious disease risk analysis method, comprising: Acquire multi-source data for the study area; The multi-source data is rasterized, and a unified grid dataset and analysis index library are constructed. The Pearson correlation coefficient between grid cases and various indicators in the unified grid dataset is calculated. Based on the Pearson coefficient, the data is sorted and screened to form a candidate influencing factor set. Factors with variance inflation factors greater than the threshold in the candidate influencing factor set are removed, and the final influencing index set is output. For each target grid cell, the neighborhood range is determined by an adaptive bandwidth method, and the spatial weights of the neighborhood samples are determined according to the specified kernel function. In each target grid cell, an exposure sequence corresponding to the current analysis period and multiple historical lag periods is constructed for each key indicator, and cross-base features are constructed. Cross-basis features constructed based on target grid cells and neighboring grid cells form a cross-basis local sample set. A geographically weighted distributed lag nonlinear model is constructed based on the cross-basis local sample set, and local regression fitting is performed. For the target grid cell, the cumulative effect value is obtained by comparing the effect differences of different exposure levels over all lag periods, and the lag period that contributes the most to the infectious disease risk of the target grid cell is determined as the optimal lag period.

[0009] According to a geographically weighted infectious disease risk analysis method provided by the present invention, multi-source data of a study area is obtained, including: Acquire population data, infectious disease case data, environmental data, population mobility data, and regional multidimensional indicator data for the study area; The population data includes the population estimate of a grid at a preset resolution provided by a specified dataset; The infectious disease case data includes the disease name, onset time, and residential address of each case; The environmental data includes meteorological data, air quality data, and land cover data; The population flow data includes the intensity of intra-city population flow calculated based on the Baidu Migration Index. The regional multidimensional indicator data includes regional GDP data, nighttime light remote sensing data, road density data, medical resource data, and public service facility data.

[0010] According to a geographically weighted infectious disease risk analysis method provided by the present invention, the multi-source data is rasterized, and a unified grid dataset and analysis indicator library are constructed, including: Multi-source data from different sources, with different spatial resolutions and different coordinate reference systems are uniformly transformed to the same coordinate reference system to obtain multi-source data with a transformed coordinate system. Construct a target grid based on the preset spatial resolution and the boundary of the study area; Multi-source data with transformed coordinate systems are mapped to the target grid to form a unified grid dataset; For continuous variables, assign values ​​using the mean, weighted mean, or interpolation method; for count variables, assign values ​​using the summation method; and for categorical variables, assign values ​​using the category with the largest area proportion. Missing values, outliers, and standardization were performed on each indicator.

[0011] According to the geographically weighted infectious disease risk analysis method provided by this invention, the Pearson correlation coefficient between grid cases and various indicators in a unified grid dataset is calculated. Based on the Pearson coefficient, candidate influencing factors are sorted and screened to form a set. Factors with variance inflation factors greater than a threshold in the candidate influencing factor set are removed, and the final set of influencing indicators is output, including: Using grid cases as the dependent variable and environmental indicators, population mobility indicators, and regional multidimensional indicators as independent variables, we extracted the values ​​of the case sequence and each indicator sequence within the corresponding period, and calculated the Pearson correlation coefficient and significance level P value between the cases and each indicator. Candidate influencing factors were obtained by removing those whose absolute value of the Pearson correlation coefficient was less than the first correlation coefficient threshold and whose significance level P value was less than the significance threshold. Calculate the Pearson correlation coefficient matrix among all candidate impact factors, and identify the candidate impact factors whose absolute Pearson correlation coefficient is greater than the second correlation coefficient threshold as highly correlated factor pairs. Variables that form highly correlated factor pairs with any factor to be tested and other candidate factors are excluded, and the remaining variables are retained as the set of predictor variables. Any factor to be tested is used as the explained variable, and a submatrix is ​​formed with the set of predictor variables to calculate the variance inflation factor value of any factor to be tested. By removing indicators whose variance inflation factor is greater than the variance inflation factor threshold, the final set of influencing indicators is obtained.

[0012] According to the geographically weighted infectious disease risk analysis method provided by the present invention, for each target grid cell, an adaptive bandwidth method is used to determine the neighborhood range, and the spatial weights corresponding to the neighborhood samples are determined according to a specified kernel function, including: Using the spatial coordinates of each target grid cell as the center, the surrounding neighboring grid cells are retrieved based on the KDTree spatial index structure; The adaptive bandwidth is determined by the distance to the preset nearest neighbor grid cell of the target grid cell, and the neighborhood range is determined by the adaptive bandwidth method. The spatial weights corresponding to the neighborhood samples are determined based on the Gaussian kernel function, and the spatial weights decrease as the distance between the neighborhood samples and the target grid cell increases.

[0013] According to the geographically weighted infectious disease risk analysis method provided by the present invention, an exposure sequence corresponding to the current analysis period and multiple historical lag periods is constructed for each key indicator in each target grid cell, and cross-basic features are constructed, including: For each target grid cell, the exposure history sequence of each key indicator is extracted within the current analysis period and several historical periods prior to it, and the maximum lag period is determined. Construct an exposure dimension spline basis for the exposure history sequence and a lag dimension spline basis for the lag periodic sequence, and determine the exposure dimension degrees of freedom, lag dimension degrees of freedom, and spline order; Cross-basis features are constructed by tensor products of exposed dimensional spline bases and hysteretic dimensional spline bases; For the current time point, cross-basis feature vectors are constructed using the exposed historical sequence, and the cross-basis outer product results under different lag periods are accumulated to form the distributed lag nonlinear design matrix column corresponding to the current grid cell and the current time point.

[0014] According to the present invention, a geographically weighted infectious disease risk analysis method is provided, which forms a cross-basis local sample set based on the cross-basis features constructed from the target grid cell and neighboring grid cells, constructs a geographically weighted distributed lag nonlinear model based on the cross-basis local sample set, and performs local regression fitting, including: For each target grid cell, the observations of multiple neighboring grid cells within its neighborhood at multiple time periods are collectively organized into a cross-basis local sample set; A time trend control term is introduced into the local sample set of the cross-basis, which is obtained by constructing a spline basis from the time series index; A geographically weighted distributed lag nonlinear model is constructed using a generalized linear model framework. When fitting the local model, the spatial weights corresponding to the neighborhood samples are introduced into the generalized linear model as frequency weights to reflect the contribution of the neighboring grid samples to the local estimation results of the target grid cell. Using all effective grid cells within the study area as the objects to be fitted, the steps of model building and local fitting are repeated sequentially or in parallel for each target grid cell.

[0015] According to the geographically weighted infectious disease risk analysis method provided by the present invention, for a target grid cell, the cumulative effect value is obtained by comparing the effect differences of different exposure levels over all lag periods, and the lag period that contributes the most to the infectious disease risk of the target grid cell is determined as the optimal lag period, including: The low-exposure and high-exposure values ​​for each influencing indicator were calculated and determined separately. Cross-basis vectors corresponding to the low-exposure and high-exposure values ​​were constructed at each lag period, and the effect difference was calculated from the regression coefficient vector obtained by local fitting. The cumulative effect value of each influencing indicator on the target grid cell is obtained by summing the effect differences over all lag periods. Calculate the relative risk of the corresponding influencing factor based on the cumulative effect value; The lag period with the largest absolute value of the effect in a single lag period is denoted as the optimal lag period; The optimal lag period of each grid cell is spatially smoothed using a weighted average smoothing method based on neighborhood kernel weights to reduce the random impact of local fluctuations in a single grid cell.

[0016] Secondly, the present invention also provides a geographically weighted infectious disease risk analysis system, comprising: The acquisition module is used to acquire multi-source data of the study area; The data processing module is used to perform rasterization processing on the multi-source data and construct a unified grid dataset and an analysis index library; the calculation module is used to calculate the Pearson correlation coefficient between grid cases and various indicators in the unified grid dataset, sort and screen according to the Pearson coefficient to form a candidate influencing factor set, remove factors with variance inflation factor greater than the threshold in the candidate influencing factor set, and output the final influencing index set. The determination module is used to determine the neighborhood range for each target grid cell using an adaptive bandwidth method and to determine the spatial weights corresponding to the neighborhood samples based on a specified kernel function. The module is used to construct the exposure sequence corresponding to the current analysis period and multiple historical lag periods for each key indicator in each target grid cell, and to construct cross-base features; The fitting module is used to form a cross-basis local sample set based on the cross-basis features constructed from the target grid cell and neighboring grid cells. Based on the cross-basis local sample set, a geographically weighted distributed lag nonlinear model is constructed, and local regression fitting is performed. The comparison module is used to obtain the cumulative effect value for the target grid cell by comparing the effect differences of different exposure levels over all lag periods, and to determine the lag period that contributes the most to the infectious disease risk of the target grid cell as the optimal lag period.

[0017] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the geographically weighted infectious disease risk analysis method as described above.

[0018] The geographically weighted infectious disease risk analysis method and system provided by this invention integrates population data, infectious disease case data, environmental data, population flow data, and regional multidimensional indicator data within a unified grid framework. It performs unified processing on data from different sources, with different spatial resolutions and coordinate reference systems, and completes rasterization construction, missing value handling, outlier handling, and standardization. This solves the problems of difficulty in coordinating and organizing multi-source data, inconsistent spatial scales, and lack of comparability between different types of influencing factors in existing technologies, thereby improving the uniformity, standardization, and refinement of infectious disease risk analysis.

[0019] This invention calculates the Pearson correlation coefficient between grid cases and various candidate indicators, and further combines the variance inflation factor to screen and eliminate candidate influencing factors. This enables the systematic screening of environmental indicators, population mobility indicators, and regional development indicators, which helps to reduce variable redundancy and collinearity interference in the model, improves the stability of parameter estimation and the interpretability of results, and overcomes the problems of existing technologies where variable selection mainly relies on empirical judgment and the screening process is unsystematic.

[0020] This invention employs a geographic weighted neighborhood construction method, which constructs a local spatiotemporal sample set on each target grid cell using its surrounding grid cells, and determines spatial weights based on adaptive bandwidth and kernel functions. This method can identify local differences in the impact of infectious disease incidence on different spatial locations, overcoming the limitations of existing technologies that use administrative regions or whole regions as modeling units and assume that different spatial locations have the same exposure-incidence relationship. This improves the ability to identify local high-risk areas and their dominant influencing factors.

[0021] This invention constructs a distributed lag nonlinear model cross-basis feature by coupling exposed dimension spline basis and lag dimension spline basis, and establishes a geographically weighted distributed lag nonlinear model under a geographically weighted framework. This model can simultaneously characterize the spatial heterogeneity, exposure nonlinearity effect and lag effect in the formation of infectious disease risk, making up for the shortcomings of existing technologies that are difficult to express the three types of features of "spatial heterogeneity, nonlinearity effect and lag effect" at the same time, and making the description of the infectious disease risk formation mechanism more comprehensive and accurate.

[0022] This invention conducts risk analysis based on gridded case data and multi-source spatial data. It does not rely on simple correlation analysis of statistical results from a single administrative region or a single exposure factor. It has the advantages of high spatial resolution, repeatable results, wide applicability and strong scalability. It can be widely applied to risk analysis, spatiotemporal early warning and public health monitoring in different regions, different types of infectious diseases and different time scales. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is one of the flowcharts of the geographically weighted infectious disease risk analysis method provided by the present invention; Figure 2 This is the second flowchart of the geographically weighted infectious disease risk analysis method provided by this invention; Figure 3This is a schematic diagram of a 3D response surface of surface temperature, air pressure, ozone, PM10 exposure and lag period, and relative risk of influenza incidence in a central urban area of ​​a certain city, provided by the present invention. Figure 4 This is a schematic diagram of the structure of the geographically weighted infectious disease risk analysis system provided by the present invention; Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] Figure 1 This is one of the flowcharts illustrating the geographically weighted infectious disease risk analysis method provided in this embodiment of the invention, such as... Figure 1 As shown, it includes: Step S1: Obtain multi-source data for the study area; Step S2: Rasterize the multi-source data and construct a unified grid dataset and analysis index library; Step S3: Calculate the Pearson correlation coefficient between grid cases and various indicators in the unified grid dataset, sort and screen according to the Pearson coefficient to form a candidate influencing factor set, remove factors with variance inflation factor greater than the threshold in the candidate influencing factor set, and output the final influencing index set; Step S4: For each target grid cell, determine the neighborhood range using an adaptive bandwidth method, and determine the spatial weights corresponding to the neighborhood samples based on the specified kernel function; Step S5: In each target grid cell, construct the exposure sequence corresponding to the current analysis period and multiple historical lag periods for each key indicator, and construct cross-base features; Step S6: Based on the cross-basis features constructed from the target grid cell and neighboring grid cells, a cross-basis local sample set is formed. A geographically weighted distributed lag nonlinear model is constructed based on the cross-basis local sample set, and local regression fitting is performed. Step S7: For the target grid cell, the cumulative effect value is obtained by comparing the effect differences of different exposure levels over all lag periods. The lag period that contributes the most to the infectious disease risk of the target grid cell is determined as the optimal lag period.

[0027] Specifically, the logic of the solution in the embodiments of the present invention is as follows: Figure 2 As shown, it includes: S1: Multi-source data acquisition: Acquiring population data, infectious disease case data, environmental data, population mobility data, and regional multidimensional indicator data for the study area. Specific details of each data item are as follows: (1) Population data: Population estimates with a 100-meter resolution grid provided by the WorldPop dataset.

[0028] (2) Infectious disease case data: The disease type is limited to influenza, including gender, age, workplace, detailed address and population classification.

[0029] (3) Environmental data include meteorological data: near-surface temperature, air pressure, absolute humidity, relative humidity, precipitation, wind speed; and air quality data: PM2.5, PM10, ozone, carbon monoxide, nitrogen dioxide, sulfur dioxide.

[0030] (4) Population flow data: The intensity of population flow within the city was further calculated based on the Baidu Migration Index. (6) Land cover data: 30-meter resolution land use data provided by the CLCD dataset and NDVI dataset created based on MODISMOD13A2 data.

[0031] (7) Regional multidimensional index data: Nighttime light remote sensing dataset, regional GDP data, road density data, medical resource data and public service facility data obtained by calibration based on DMSP-OLS data.

[0032] S2: Multi-source data rasterization and unified grid construction: Rasterization of the various types of data obtained in step S1 is performed, and a unified target grid dataset and analysis indicator library are constructed. Specifically, this includes: (1) The geographic coordinate reference system of the population data, meteorological data, air quality data, land cover data and regional multidimensional index data obtained in step S1 is unified to the WGS-1984 coordinate system through projection transformation.

[0033] (2) The preset spatial resolution is 1km×1km. The grid distribution after resampling the population data grid to the preset spatial resolution is the target grid shape.

[0034] (3) For the infectious disease case data obtained in step S1, according to the detailed address of each case, use Baidu Map Geocoding API service to geocode the detailed address of the case to the corresponding longitude and latitude and map it to the corresponding target grid.

[0035] (4) The case data on the grid is summed to obtain a long-term case grid dataset. The precipitation raster dataset is resampled by summing to obtain a precipitation dataset that conforms to the target grid. The meteorological data, air quality data, NDVI data and socio-economic development data other than precipitation are resampled by mean to obtain a dataset that conforms to the target grid. The land use data is assigned and resampled by the largest area proportion category to obtain a dataset that conforms to the target grid.

[0036] (5) Perform missing value processing, outlier processing and standardization processing on each indicator.

[0037] S3: Screening of Key Influencing Indicators: Calculate the Pearson correlation coefficient between gridded cases and each indicator, and further calculate the variance inflation factor of the candidate influencing factor set. Eliminate indicators with a variance inflation factor greater than 10 to obtain the final set of influencing indicators used in subsequent modeling. Specific steps are as follows: (1) Using the grid case value as the dependent variable and the environmental indicators, population mobility indicators and regional multidimensional development indicators as independent variables, the values ​​of the case sequence and each indicator sequence within the corresponding period were taken respectively, and the Pearson correlation coefficient and significance level P value between the case and each indicator were calculated respectively.

[0038] (2) Remove the indicators with an absolute value of correlation coefficient less than 0.1 and a significance level P value less than 0.01, and retain the candidate influencing factors with strong correlation.

[0039] (3) For all candidate impact factors, calculate the Pearson correlation coefficient matrix between the candidate impact factors, and consider the factor pairs with a Pearson correlation coefficient greater than 0.6 as highly correlated factor pairs.

[0040] (4) Continue to calculate the variance inflation factor for the candidate influencing factor set. For a certain factor to be tested, exclude the variables that form a highly correlated factor pair with the factor from the remaining candidate factors and retain only the remaining variables as the predictor variable set. Then, take the factor to be tested as the explained variable and form a submatrix with the predictor variable set to calculate the variance inflation factor (VIF) value of the factor.

[0041] (5) Remove indicators with variance inflation factor greater than 10 to eliminate the severe collinearity effect, thereby obtaining the final set of influencing indicators.

[0042] S4: Geographically Weighted Neighborhood Construction and Spatial Weight Determination: For each target grid cell, an adaptive bandwidth method is used to determine the neighborhood range; the spatial weight of the neighborhood sample is determined based on the Gaussian kernel function, and the spatial weight of the neighborhood sample decreases as the distance between the neighborhood sample and the target grid cell increases. Specifically, this includes: (1) Using the spatial coordinates of each target grid cell as the center, retrieve its surrounding neighboring grid cells based on the KDTree spatial index structure.

[0043] (2) The neighborhood range is determined by the adaptive bandwidth method. The adaptive bandwidth is determined by the distance to the 20th nearest neighbor grid cell of the target grid cell. (3) Determine the spatial weights corresponding to the neighborhood samples based on the Gaussian kernel function. The spatial weights decrease as the distance between the neighborhood samples and the target grid cell increases. The Gaussian kernel function can be expressed as:

[0044] in Represents the spatial weights of neighboring samples. This represents the distance between neighboring samples and the target grid cell. Indicates bandwidth.

[0045] S5: Construction of Cross-Basis Features for Distributed Lag Nonlinear Model: In each grid cell, exposure sequences corresponding to the current analysis period and multiple historical lag periods are constructed for each key indicator, and cross-basis features are constructed as follows: (1) For each target grid cell, extract the exposure history sequence of each key indicator in the current analysis period and several previous historical periods, with the maximum lag period set to 8 periods.

[0046] (2) Construct an exposed dimension spline basis for the exposed sequence and a lag dimension spline basis for the lag periodic sequence; wherein the exposed dimension degrees of freedom are set to 4, the lag dimension degrees of freedom are set to 4, and the spline order is set to 3.

[0047] (3) Construct cross-basis features by tensor product of exposed-dimensional spline basis and hysteretic-dimensional spline basis. For any time point and the lag period If the exposed dimensional spline basis is denoted as The lag dimension spline basis is denoted as Then the cross basis can be represented as:

[0048] in This represents the tensor product operation. Indicates exposure value In the lag period The corresponding cross-base features.

[0049] For the current time point Using historical exposure sequences Construct cross-basis eigenvectors and accumulate the cross-basis outer product results under different lag periods to form the distributed lag nonlinear design matrix column corresponding to the current grid cell and the current time point.

[0050] S6: Construction and Local Fitting of Geographically Weighted Distributed Lag Nonlinear Model: A cross-basis local sample set is constructed using the target grid cell and its neighboring grid cells to build the corresponding geographically weighted distributed lag nonlinear model, and local regression fitting is performed. Specifically, this includes: (1) Based on the neighborhood grid cells determined in step S4, for each target grid cell, the observations of multiple grid cells in its neighborhood over multiple time periods are jointly organized into a local sample set. Specifically, for each sample record in the local sample set, a cross-basis feature vector is constructed for each key influence indicator, and the cross-basis vectors corresponding to all indicators are concatenated column-wise to form the core design matrix of the local model. The cross-basis feature of a single influence indicator is obtained by expanding the tensor product of the exposed dimension spline basis and the lag dimension spline basis. If the degrees of freedom of the exposed dimension of each variable are... The lag dimension has degrees of freedom of Then the number of cross basis columns generated by a single variable is .

[0051] (2) In order to reduce the interference of long-term trends, seasonal fluctuations and other time-varying but unobserved factors on the estimation of local exposure-pathogenesis relationship, a time trend control term is introduced into the local model. Specifically, the time trend term is obtained by constructing a spline basis from the time series index, with its degrees of freedom set to 6, and a constant term is added to the time base matrix to simultaneously represent the time smoothing trend and the model intercept.

[0052] (3) A geographically weighted distributed lag nonlinear model is constructed using a generalized linear model framework. Specifically, for the target grid cell... Let the first local sample set be the... The response variable corresponding to each sample is , representing the number of infectious disease cases at the corresponding spatiotemporal location of the sample; let its expected value be . Therefore, the following locally geographically weighted distributed lag nonlinear model is established:

[0053] in Represents the target grid cell The intercept term corresponds to the local model; P represents the number of key impact indicators included in the model; Indicates the first The number of columns of cross-base features corresponding to each influencing indicator; Indicates the first In the nth sample, the nth The first influencing indicator corresponds to the first Each cross-base eigenvalue; Represents the target grid cell Place, No. The first influencing indicator Cross-base coefficients; Indicates the first Each sample at time point The first The values ​​of the time spline basis function; Represents the target grid cell The regression coefficient of the corresponding time trend term; This represents the number of time spline basis functions.

[0054] (4) During local model fitting, the spatial weights corresponding to the neighborhood samples are introduced into the generalized linear model as frequency weights to reflect the contribution of neighboring grid samples to the local estimation results of the target grid cell. Using the spatial weights obtained in step S4, for the target grid cell... The local sample set, the first The weights of each sample are then... The parameter estimation for the aforementioned locally weighted generalized linear model can be solved using the weighted likelihood function, which is expressed as follows:

[0055] in Represents the target grid cell The parameter vector corresponding to the local model; Represents the target grid cell Corresponding local sample number; Indicates the first The actual number of cases in a local sample; Indicates the first The expected value of the model for a local sample.

[0056] (5) Using all effective grid cells within the study area as the objects to be fitted, repeat the processing steps (1) to (4) for each target grid cell sequentially or in parallel. Since the neighborhood composition, spatial weights, and local sample distributions corresponding to different target grid cells are all different, a set of local model parameters that vary with spatial location can be obtained in the end.

[0057] S7: Cumulative Effect and Optimal Lag Period Calculation: For the target grid cell, the cumulative effect value is obtained by comparing the difference in effects between high and low exposure levels over all lag periods, and the optimal lag period is further determined, specifically including: (1) For each influencing indicator, calculate the exposure values ​​corresponding to its low and high exposure levels, where the low exposure level uses the 10th percentile and the high exposure level uses the 90th percentile. In each lag period... Above, construct cross-basis vectors corresponding to high and low exposure values ​​respectively, and calculate the effect difference between them; let the th... The difference in effect over each lag period is Then it can be expressed as:

[0058] in, This indicates high exposure levels over a lag period. The corresponding cross basis vectors, This indicates low exposure levels over a lag period. The corresponding cross basis vectors, This represents the regression coefficient vector obtained from local fitting.

[0059] (2) The cumulative effect value of the influencing index on the target grid cell is obtained by summing the effect differences over all lag periods. The calculation formula is as follows:

[0060] (3) Furthermore, the relative risk (RR) of the corresponding influencing factor can be calculated based on the cumulative effect value, as shown in the figure. Figure 3 As shown, the formula is as follows:

[0061] (4) The lag period with the largest absolute value of the effect in a single lag period is denoted as the optimal lag period, and its determination formula is as follows:

[0062] (5) The optimal lag period results of each grid cell are further processed by spatial smoothing using weighted average smoothing based on neighborhood kernel weights to reduce the random impact of local fluctuations in a single grid cell.

[0063] The geographically weighted infectious disease risk analysis system provided by this invention is described below. The geographically weighted infectious disease risk analysis system described below can be referred to in correspondence with the geographically weighted infectious disease risk analysis method described above.

[0064] Figure 4 This is a schematic diagram of the structure of the geographically weighted infectious disease risk analysis system provided in an embodiment of the present invention, as shown below. Figure 4 As shown, it includes: an acquisition module 41, a data processing module 42, a calculation module 43, a determination module 44, a construction module 45, a fitting module 46, and a comparison module 47, wherein: The acquisition module 41 is used to acquire multi-source data of the study area; the data processing module 42 is used to perform rasterization processing on the multi-source data and construct a unified grid dataset and an analysis index library; the calculation module 43 is used to calculate the Pearson correlation coefficient between grid cases and various indicators in the unified grid dataset, sort and screen according to the Pearson coefficient to form a candidate influencing factor set, remove factors with variance inflation factors greater than a threshold in the candidate influencing factor set, and output the final influencing index set; the determination module 44 is used to determine the neighborhood range for each target grid cell using an adaptive bandwidth method, and determine the spatial weights corresponding to the neighborhood samples according to a specified kernel function. The module 45 is used to construct the exposure sequence corresponding to the current analysis period and multiple historical lag periods for each key indicator in each target grid cell, and to construct cross-base features; the fitting module 46 is used to form a cross-base local sample set based on the cross-base features constructed based on the target grid cell and neighboring grid cells, construct a geographically weighted distributed lag nonlinear model based on the cross-base local sample set, and perform local regression fitting; the comparison module 47 is used to obtain the cumulative effect value for the target grid cell by comparing the effect differences of different exposure levels over all lag periods, and determine the lag period that contributes the most to the infectious disease risk of the target grid cell as the optimal lag period.

[0065] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540. The processor 510, communication interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a geographically weighted infectious disease risk analysis method. This method includes: acquiring multi-source data of the study area; rasterizing the multi-source data and constructing a unified grid dataset and an analysis indicator library; calculating the Pearson correlation coefficient between grid cases and various indicators in the unified grid dataset; sorting and screening based on the Pearson coefficient to form a candidate influencing factor set; removing factors with variance inflation factors greater than a threshold from the candidate influencing factor set; and outputting the final influencing indicator set; for each target grid cell, determining the neighborhood range using an adaptive bandwidth method. The spatial weights of neighboring samples are determined based on a specified kernel function. For each target grid cell, exposure sequences corresponding to the current analysis period and multiple historical lag periods are constructed for each key indicator, and cross-basis features are built. A cross-basis local sample set is formed based on the cross-basis features constructed from the target grid cell and neighboring grid cells. A geographically weighted distributed lag nonlinear model is constructed based on the cross-basis local sample set, and local regression fitting is performed. For the target grid cell, the cumulative effect value is obtained by comparing the effect differences of different exposure levels over all lag periods, and the lag period that contributes the most to the infectious disease risk of the target grid cell is determined as the optimal lag period.

[0066] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0067] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0068] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A geographically weighted method for infectious disease risk analysis, characterized in that, include: Acquire multi-source data for the study area; The multi-source data is rasterized, and a unified grid dataset and analysis index library are constructed. The Pearson correlation coefficient between grid cases and various indicators in the unified grid dataset is calculated. Based on the Pearson coefficient, the data is sorted and screened to form a candidate influencing factor set. Factors with variance inflation factors greater than the threshold in the candidate influencing factor set are removed, and the final influencing index set is output. For each target grid cell, the neighborhood range is determined by an adaptive bandwidth method, and the spatial weights of the neighborhood samples are determined according to the specified kernel function. In each target grid cell, an exposure sequence corresponding to the current analysis period and multiple historical lag periods is constructed for each key indicator, and cross-base features are constructed. Cross-basis features constructed based on target grid cells and neighboring grid cells form a cross-basis local sample set. A geographically weighted distributed lag nonlinear model is constructed based on the cross-basis local sample set, and local regression fitting is performed. For the target grid cell, the cumulative effect value is obtained by comparing the effect differences of different exposure levels over all lag periods, and the lag period that contributes the most to the infectious disease risk of the target grid cell is determined as the optimal lag period.

2. The geographically weighted infectious disease risk analysis method according to claim 1, characterized in that, Acquire multi-source data for the study area, including: Acquire population data, infectious disease case data, environmental data, population mobility data, and regional multidimensional indicator data for the study area; The population data includes the population estimate of a grid at a preset resolution provided by a specified dataset; The infectious disease case data includes the disease name, onset time, and residential address of each case; The environmental data includes meteorological data, air quality data, and land cover data; The population flow data includes the intensity of intra-city population flow calculated based on the Baidu Migration Index. The regional multidimensional indicator data includes regional GDP data, nighttime light remote sensing data, road density data, medical resource data, and public service facility data.

3. The geographically weighted infectious disease risk analysis method according to claim 1, characterized in that, The multi-source data is rasterized, and a unified grid dataset and analysis indicator library are constructed, including: Multi-source data from different sources, with different spatial resolutions and different coordinate reference systems are uniformly transformed to the same coordinate reference system to obtain multi-source data with a transformed coordinate system. Construct a target grid based on the preset spatial resolution and the boundary of the study area; Multi-source data with transformed coordinate systems are mapped to the target grid to form a unified grid dataset; For continuous variables, assign values ​​using the mean, weighted mean, or interpolation method; for count variables, assign values ​​using the summation method; and for categorical variables, assign values ​​using the category with the largest area proportion. Missing values, outliers, and standardization were performed on each indicator.

4. The geographically weighted infectious disease risk analysis method according to claim 1, characterized in that, Calculate the Pearson correlation coefficient between grid cases and various indicators in the unified grid dataset. Based on the Pearson coefficient, sort and filter to form a candidate influencing factor set. Remove factors from the candidate influencing factor set whose variance inflation factor is greater than a threshold. Output the final influencing indicator set, including: Using grid cases as the dependent variable and environmental indicators, population mobility indicators, and regional multidimensional indicators as independent variables, we extracted the values ​​of the case sequence and each indicator sequence within the corresponding period, and calculated the Pearson correlation coefficient and significance level P value between the cases and each indicator. Candidate influencing factors were obtained by removing those whose absolute value of the Pearson correlation coefficient was less than the first correlation coefficient threshold and whose significance level P value was less than the significance threshold. Calculate the Pearson correlation coefficient matrix among all candidate impact factors, and identify the candidate impact factors whose absolute Pearson correlation coefficient is greater than the second correlation coefficient threshold as highly correlated factor pairs. Variables that form highly correlated factor pairs with any factor to be tested and other candidate factors are excluded, and the remaining variables are retained as the set of predictor variables. Any factor to be tested is used as the explained variable, and a submatrix is ​​formed with the set of predictor variables to calculate the variance inflation factor value of any factor to be tested. By removing indicators whose variance inflation factor is greater than the variance inflation factor threshold, the final set of influencing indicators is obtained.

5. The geographically weighted infectious disease risk analysis method according to claim 1, characterized in that, For each target grid cell, an adaptive bandwidth method is used to determine the neighborhood range, and the spatial weights corresponding to the neighborhood samples are determined according to a specified kernel function, including: Using the spatial coordinates of each target grid cell as the center, the surrounding neighboring grid cells are retrieved based on the KDTree spatial index structure; The adaptive bandwidth is determined by the distance to the preset nearest neighbor grid cell of the target grid cell, and the neighborhood range is determined by the adaptive bandwidth method. The spatial weights corresponding to the neighborhood samples are determined based on the Gaussian kernel function, and the spatial weights decrease as the distance between the neighborhood samples and the target grid cell increases.

6. The geographically weighted infectious disease risk analysis method according to claim 1, characterized in that, In each target grid cell, exposure sequences corresponding to the current analysis period and multiple historical lag periods are constructed for each key indicator, and cross-basis features are constructed, including: For each target grid cell, the exposure history sequence of each key indicator is extracted within the current analysis period and several historical periods prior to it, and the maximum lag period is determined. Construct an exposure dimension spline basis for the exposure history sequence and a lag dimension spline basis for the lag periodic sequence, and determine the exposure dimension degrees of freedom, lag dimension degrees of freedom, and spline order; Cross-basis features are constructed by tensor products of exposed dimensional spline bases and hysteretic dimensional spline bases; For the current time point, cross-basis feature vectors are constructed using the exposed historical sequence, and the cross-basis outer product results under different lag periods are accumulated to form the distributed lag nonlinear design matrix column corresponding to the current grid cell and the current time point.

7. The geographically weighted infectious disease risk analysis method according to claim 1, characterized in that, Cross-basis features constructed based on target grid cells and neighboring grid cells form a cross-basis local sample set. A geographically weighted distributed lag nonlinear model is constructed based on this local sample set, and local regression fitting is performed, including: For each target grid cell, the observations of multiple neighboring grid cells within its neighborhood at multiple time periods are collectively organized into a cross-basis local sample set; A time trend control term is introduced into the local sample set of the cross-basis, which is obtained by constructing a spline basis from the time series index; A geographically weighted distributed lag nonlinear model is constructed using a generalized linear model framework. When fitting the local model, the spatial weights corresponding to the neighborhood samples are introduced into the generalized linear model as frequency weights to reflect the contribution of the neighboring grid samples to the local estimation results of the target grid cell. Using all effective grid cells within the study area as the objects to be fitted, the steps of model building and local fitting are repeated sequentially or in parallel for each target grid cell.

8. The geographically weighted infectious disease risk analysis method according to claim 1, characterized in that, For the target grid cell, the cumulative effect value is obtained by comparing the differences in the effects of different exposure levels over all lag periods. The lag period that contributes the most to the infectious disease risk of the target grid cell is determined as the optimal lag period, including: The low-exposure and high-exposure values ​​for each influencing indicator were calculated and determined separately. Cross-basis vectors corresponding to the low-exposure and high-exposure values ​​were constructed at each lag period, and the effect difference was calculated from the regression coefficient vector obtained by local fitting. The cumulative effect value of each influencing indicator on the target grid cell is obtained by summing the effect differences over all lag periods. Calculate the relative risk of the corresponding influencing factor based on the cumulative effect value; The lag period with the largest absolute value of the effect in a single lag period is denoted as the optimal lag period; The optimal lag period of each grid cell is spatially smoothed using a weighted average smoothing method based on neighborhood kernel weights to reduce the random impact of local fluctuations in a single grid cell.

9. A geographically weighted infectious disease risk analysis system, characterized in that, include: The acquisition module is used to acquire multi-source data of the study area; The data processing module is used to perform rasterization processing on the multi-source data and construct a unified grid dataset and an analysis index library; the calculation module is used to calculate the Pearson correlation coefficient between grid cases and various indicators in the unified grid dataset, sort and screen according to the Pearson coefficient to form a candidate influencing factor set, remove factors with variance inflation factor greater than the threshold in the candidate influencing factor set, and output the final influencing index set. The determination module is used to determine the neighborhood range for each target grid cell using an adaptive bandwidth method and to determine the spatial weights corresponding to the neighborhood samples based on a specified kernel function. The module is used to construct the exposure sequence corresponding to the current analysis period and multiple historical lag periods for each key indicator in each target grid cell, and to construct cross-base features; The fitting module is used to form a cross-basis local sample set based on the cross-basis features constructed from the target grid cell and neighboring grid cells. Based on the cross-basis local sample set, a geographically weighted distributed lag nonlinear model is constructed, and local regression fitting is performed. The comparison module is used to obtain the cumulative effect value for the target grid cell by comparing the effect differences of different exposure levels over all lag periods, and to determine the lag period that contributes the most to the infectious disease risk of the target grid cell as the optimal lag period.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the geographically weighted infectious disease risk analysis method as described in any one of claims 1 to 8.

Images