Child respiratory tract infection risk prediction method based on multi-source medical data

Abstract

The invention discloses a children respiratory tract infection risk prediction method based on multi-source medical data, particularly relates to the field of respiratory tract infection risk prediction, and is used for solving the problems that an existing method cannot identify a cross-regional co-infection mode and cannot prevent an input transmission risk. According to the method, a high-incidence time period baseline is constructed by extracting children respiratory tract infection historical electronic medical records of all regions, a local co-infection mode feature library is established based on a pathogen positive case co-occurrence relation, and a region-time period-pathogen three-dimensional collaborative tensor is constructed to recognize a synchronous high-incidence region group. And capturing real-time pathogen data of the synchronous high-incidence area, matching the real-time pathogen data with a local co-infection mode to generate a real-time risk list, deducing an input co-infection mode in combination with interregional personnel circulation intensity, and calculating the cross-regional risk permeability. And finally, a risk heat map is generated based on the permeability, and regional linkage-oriented children respiratory tract infection risk prediction is realized.

Description

Technical Field

[0001] This invention relates to the field of respiratory infection risk prediction technology, and more specifically, to a method for predicting the risk of respiratory infections in children based on multi-source medical data. Background Technology

[0002] Respiratory infections in children consistently account for a high proportion of daily outpatient visits in medical institutions across various regions, especially during the autumn and winter seasons and in urban clusters with frequent population movement. Case numbers often exhibit complex characteristics such as sudden increases, synchronous fluctuations across regions, and rapid changes in the pathogen spectrum. With the combined effects of living environment, climate conditions, concentrated school activities, inter-provincial travel, and community transmission, it has become increasingly common for different regions to experience high infection rates driven by multiple pathogens within similar timeframes. Furthermore, historical trend analysis of a single region is no longer sufficient to effectively depict common infection patterns across regions.

[0003] Furthermore, as respiratory pathogens exhibit trends of co-occurrence, multiple superimposed infections, and spectrum evolution, monitoring of a single pathogen alone is insufficient to support regional-level forecasting needs. This is especially true in the context of large-scale two-way population flows between urban clusters, where pathogens spread rapidly and infiltrate covertly between regions. Medical institutions often only passively discover associated risks after a significant increase in cases.

[0004] Existing technologies lack a risk prediction method that can integrate multi-source medical data, correlate cross-regional spatiotemporal transmission processes, and dynamically identify imported co-infection patterns. They cannot perform fusion analysis based on changes in pathogen spectrum, real-time detection data, and inter-regional flow intensity, nor can they provide forward-looking predictions before cross-regional risk penetration occurs. This makes it difficult to meet the current practical needs of children's respiratory infection prevention and control for high timeliness and regional coordination. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a method for predicting the risk of respiratory tract infection in children based on multi-source medical data to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting the risk of respiratory infections in children based on multi-source medical data includes the following steps: S1. Extract historical electronic medical records of children's respiratory infections and pathogen detection data from local medical institutions in various regions to construct a baseline for the high incidence period of children's respiratory infections in the local area; S2. Perform time-segment analysis on historical pathogen-positive cases in various regions to identify pathogen co-occurrence combinations during each peak period of respiratory infection and construct a local co-infection pattern feature library. S3. Integrate the baseline of high-incidence periods in all regions, establish a three-dimensional collaborative matrix of regions, high-incidence periods and pathogens, and identify synchronous high-incidence regions with different pathogen spectra through spatiotemporal mode decomposition; S4. Capture real-time pathogen detection data from each region within the synchronous high-incidence area group, match it with the corresponding local co-infection pattern feature library, and generate a real-time co-infection risk list. S5. Using high-incidence areas as units, cross-compare the real-time risk lists of each region, combine the intensity of inter-regional personnel flow, deduce the imported co-infection pattern, and calculate the cross-regional risk penetration rate. S6. Based on the imported co-infection pattern and the corresponding cross-regional risk penetration rate, establish a visualized dynamic risk heat map.

[0007] As a further aspect of the present invention, in S1, extracting historical electronic medical records of children's respiratory infections and pathogen detection data from local medical institutions in various regions to construct a baseline for high-incidence periods of local children's respiratory infections specifically includes: The pathogen detection data includes target pathogen types, detection frequency, and detection result data within a preset range; Extract pediatric respiratory infection data from historical electronic medical records in time series, calculate the moving average ratio of the number of cases in each time period to the number of cases in adjacent time periods, and use it as an indicator of local activity. The local activity index is combined with the historical average number of cases in the same period to establish a comprehensive time-series feature vector; Density clustering algorithm was used to analyze the distribution of the comprehensive temporal feature vector on the time axis, identify continuous time periods with feature vector density higher than the average level, and integrate them into a baseline of high-incidence periods of respiratory infections in local children.

[0008] As a further aspect of the present invention, in S2, performing time-segment analysis on historical pathogen-positive cases in various regions to identify pathogen co-occurrence combinations within each high-incidence period of respiratory infection and constructing a local co-infection pattern feature library specifically includes: Extract the high-incidence time periods from the baseline of high-incidence periods in each region, aggregate all historical pathogen detection data with positive test results within each high-incidence time period, and mark the corresponding cases as positive cases; Using association rule mining algorithms, we analyze frequent itemsets in each batch of positive cases where multiple pathogens co-occur, and calculate the joint support of each itemset. The pattern strength index is calculated based on the ratio of joint support to the independent support of each pathogen in the itemset, and pathogen co-occurrence combinations with pattern strength indices exceeding a preset threshold are screened. Pathogen co-occurrence combinations are stored in a feature library, a set of co-infection patterns for high-incidence time periods is established, and all sets of high-incidence time periods are aggregated to form a local co-infection pattern feature library.

[0009] As a further aspect of the present invention, in S3, integrating the baselines of high-incidence periods in all regions to establish a three-dimensional collaborative matrix of regions, high-incidence periods, and pathogens, and identifying synchronously high-incidence regions with differences in pathogen spectra through spatiotemporal mode decomposition specifically includes: Extract the high-incidence time periods from the baseline of high-incidence periods in each region, and the detection frequency of different types of single pathogens in the corresponding historical pathogen detection data, construct a pathogen spectrum and label the dominant pathogen set; The baseline of high-incidence periods in all regions is collected, and a three-dimensional collaborative tensor with region, high-incidence period and pathogen type as the axis is constructed. Spatiotemporal constraint clustering is performed on the three-dimensional collaborative tensor to extract the initial cluster of synchronous high-incidence regions. The spatiotemporal constraints include the synchronicity constraint of high-incidence period and the spatial connectivity constraint in the region dimension. Calculate the pathogen spectrum difference between any two regions within the initial cluster of regions based on the dominant pathogen set, where the difference is the proportion of the number of asymmetrically different pathogens in the dominant pathogen sets of the two regions to the total number of pathogens in the union set. Clusters in which the mean difference in pathogen spectrum between all regions within the initial cluster of the selected region is greater than a set value are labeled as synchronously high-incidence regions with significant differences.

[0010] As a further aspect of the present invention, the value of each element in the three-dimensional cooperative tensor is a binary identifier, which respectively indicates whether the pathogen is listed as the dominant pathogen in its pathogen spectrum characteristics in the corresponding region during the corresponding high-incidence period.

[0011] As a further aspect of the present invention, in step S4, capturing real-time pathogen detection data from various regions within the synchronous high-incidence area group and matching it with the corresponding local co-infection pattern feature library to generate a real-time co-infection risk list specifically includes: Real-time extraction of pathogen-positive case data for each region within the synchronous high-incidence area group, generating a current deduplicated real-time pathogen type set for each region; The real-time pathogen set is compared with the pathogen co-occurrence combination of each co-infection pattern unit in the local co-infection pattern feature library of the corresponding region. The pathogen co-occurrence combination with intersection is used as a risk item, and the pathogens that are not hit in the pathogen co-occurrence combination are marked as permeable pathogens. All selected risk items are aggregated to generate a real-time co-infection risk list for each region.

[0012] As a further aspect of the present invention, in S5, taking high-incidence areas as units, cross-referencing the real-time risk lists of each region, and combining the intensity of inter-regional personnel flow to deduce the imported co-infection pattern and calculate the cross-regional risk penetration rate specifically includes: Extract the real-time co-infection risk list for each region within the high-incidence region group. For each pair of regions with personnel flow within the group, pair the pathogens marked as permeable in the risk list of the destination region with the existing pathogen co-occurrence combinations in the risk list of the source region to generate potential risk pathogen co-occurrence combinations in the destination region activated by adding permeable pathogens from the source region to the original pathogen co-occurrence combinations in the risk list of the source region. Co-occurrence combinations of potentially risky pathogens are considered as imported co-infection patterns; The cross-regional risk penetration rate of potential risk pathogen co-occurrence combinations in the destination area is calculated by multiplying the intensity of personnel flow from the source area to the destination area by the dimensionless product of the frequency of occurrence of the permeable pathogen in all risk items of the risk list in the destination area. Using the same method, pathogens marked as permeable in the source region risk list are reverse-paired with existing pathogen co-occurrence combinations in the destination region risk list, and the cross-regional risk penetration rate is calculated.

[0013] As a further aspect of the present invention, in step S6, establishing a visualized dynamic risk heatmap based on imported co-infection patterns and corresponding cross-regional risk penetration rates specifically includes: By aggregating potential risk pathogen co-occurrence combinations generated from all directions and their corresponding cross-regional risk penetration rates, a two-way imported co-infection pattern is formed and updated in real time, establishing a visualized dynamic risk heat map of respiratory infections in children.

[0014] The technical effects and advantages of the present invention, a method for predicting the risk of respiratory infections in children based on multi-source medical data, are as follows: By deeply integrating historical electronic medical records, pathogen detection data, and cross-regional spatiotemporal characteristics, this method enables the prospective identification of respiratory infection risks in children and the dynamic extrapolation of cross-regional transmission paths. Compared to traditional methods relying on single-region monitoring or single-pathogen trend analysis, its technological advantages lie in its ability to systematically characterize regional high-incidence periods, co-infection patterns, and pathogen spectrum differences, and to automatically identify imported co-infection patterns in real-world scenarios with inter-regional population movement. By constructing a three-dimensional collaborative matrix of regions, high-incidence periods, and pathogens, it can identify synchronously high-incidence regional groups that are difficult to identify using traditional time-series models. Combined with a mechanism for calculating the penetration rate of permeable pathogens and cross-regional risk penetration rates, the early warning results possess stronger cross-regional interpretability and spatiotemporal linkage. The combination of real-time risk lists and imported co-infection patterns allows medical institutions in various regions to know the possible pathogen combinations before the risk penetration arrives, thereby assisting in the formulation of resource allocation, testing reagent reserves, and key population monitoring strategies. Finally, dynamic risk heatmaps present the complex transmission process as visualized spatial risk information, improving the intuitiveness and real-time nature of regional respiratory infection risk assessment. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a method for predicting the risk of respiratory infections in children based on multi-source medical data according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 Figure 1 This invention presents a method for predicting the risk of respiratory infections in children based on multi-source medical data, which includes the following steps: S1. Extract historical electronic medical records of children's respiratory infections and pathogen detection data from local medical institutions in various regions to construct a baseline for the high incidence period of children's respiratory infections in the local area; S2. Perform time-segment analysis on historical pathogen-positive cases in various regions to identify pathogen co-occurrence combinations during each peak period of respiratory infection and construct a local co-infection pattern feature library. S3. Integrate the baseline of high-incidence periods in all regions, establish a three-dimensional collaborative matrix of regions, high-incidence periods and pathogens, and identify synchronous high-incidence regions with different pathogen spectra through spatiotemporal mode decomposition; S4. Capture real-time pathogen detection data from each region within the synchronous high-incidence area group, match it with the corresponding local co-infection pattern feature library, and generate a real-time co-infection risk list. S5. Using high-incidence areas as units, cross-compare the real-time risk lists of each region, combine the intensity of inter-regional personnel flow, deduce the imported co-infection pattern, and calculate the cross-regional risk penetration rate. S6. Based on the imported co-infection pattern and the corresponding cross-regional risk penetration rate, establish a visualized dynamic risk heat map.

[0018] In S1, the electronic medical records of children's respiratory tract infection history and pathogen detection data of local medical institutions in each region are extracted to construct a baseline for the high incidence period of children's respiratory tract infection in the local area.

[0019] The pathogen detection data explicitly includes a predefined range of target pathogen types, detection frequencies, and corresponding test results. The predefined range of target pathogen types is set by combining typical causative agents of respiratory infections in children, ensuring the stability and clarity of the target pathogen range and avoiding uncontrollable data drift during processing. Common respiratory pathogens are included in the predefined categories as a list of target pathogen types. Each pathogen corresponds to specific test results, such as a positive nucleic acid test or a positive antigen test, combined with detection frequency information in the testing process, i.e., the percentage of times that type of pathogen is detected in all tested samples within the same time period. Test result data is obtained from the hospital's recorded laboratory testing logs. Based on the above target pathogen types and detection frequencies, a consistent and stable pathogen detection dataset is formed within a specific time span, providing a clear data structure foundation for subsequent analysis. Furthermore, if the hospital's testing capabilities cover more pathogen types during implementation, they can also be included in the predefined range, as long as the pathogen type field remains consistent during the data import process.

[0020] By analyzing pediatric respiratory infection case data from historical electronic medical records (EMRs) over time periods, a case count sequence on a continuous timeline was created. Historical EMRs not only record the infection events themselves but also contain fundamental data such as patient age, date of visit, main symptoms, and diagnosis type. Cases of children under 12 years old were filtered by age. To ensure the accuracy of the time series analysis, statistics were collected at a fixed time granularity, such as by day, week, or finer time windows. This time granularity was set and maintained consistently before project deployment. Subsequently, by observing the trend of case count changes within continuous time periods, the ratio of the number of cases in each time period to the moving average of the number of cases in adjacent time periods was calculated. This ratio directly characterizes the growth or contraction trend of case counts within a local time window, thus forming a local activity index. The local activity index was combined with the historical average number of cases for the same period to simultaneously reflect both the normalized case level and the current actual case fluctuations within the same season or public health context. The combination method involves splicing the two indicators at a fixed ratio to form a single vector, giving it complete expressive power. Ultimately, the combination of indicators across all time periods forms a continuously arranged sequence of comprehensive time-series feature vectors.

[0021] Using the constructed comprehensive temporal feature vector sequence as input data, and with time as the sole analysis axis, density clustering analysis is performed on the entire temporal vector distribution. In this scenario, the density clustering algorithm is used to identify continuous time periods with concentrated feature distributions within the time dimension, i.e., high-activity intervals with significantly increased case activity. First, the comprehensive temporal feature vector is embedded into the timeline, and the local density at each point is estimated by setting a density threshold. The density threshold is set based on the average distribution characteristics of the vector in the historical sequence; for example, a percentile slightly higher than the overall density mean is selected as the threshold to accurately distinguish between normal fluctuations and abnormally high-density time periods. Second, by continuously scanning the timeline, all time points with densities higher than the threshold are marked, forming multiple initial segments. To ensure the continuity and interpretability of high-incidence periods, initial segments with small adjacent distances are merged to form a set of high-incidence time periods with stable spans. During the merging process, a fixed time interval is used as the criterion; for example, if the interval between two segments does not exceed a preset value, they are considered as the same whole segment. Furthermore, segments with very short spans but extremely high densities can also be retained as independent high-incidence segments. Ultimately, all the data were integrated into a baseline for the peak periods of respiratory infections in local children, and recorded in a clearly defined time interval format.

[0022] In S2, time-segment analysis is performed on historical pathogen-positive cases in various regions to identify pathogen co-occurrence combinations during each high-incidence period of respiratory infection and to construct a local co-infection pattern feature library.

[0023] Based on the baseline of high-incidence periods identified in each region, the corresponding high-incidence time periods for each region are extracted and fixed as specific time intervals. Then, using the high-incidence time periods as a time filter, historical pathogen testing data recorded by hospital laboratories are screened, extracting all records with positive test results within that time period. Test results include clearly defined positive markers such as positive nucleic acid test, positive antigen detection, or positive immunological test for various pathogens. All positive records, regardless of the department, batch, or instrument used, are included in the aggregation scope as long as they exist within the predefined list of target pathogen types. To ensure a clear data structure, each positive record is associated with its original case number, so that multiple test results for each case within the high-incidence time period are categorized by the same case identifier. If the same case tests positive for multiple pathogens within the same time period, all these positive results are retained and associated with the same case to fully present the co-occurrence of pathogens. Furthermore, for records with identical results from repeated tests on the same day, deduplication is performed according to hospital rules, such as retaining the first positive record and removing other duplicate records to avoid counting bias. After completing the above aggregation, a list of positive pathogens for all cases during the high-incidence period is generated, and each case is marked as a positive case. After aggregating positive cases during peak periods, these cases are batched according to their respective peak periods, forming a case set structure that can be input into the association rule mining algorithm. Each case set contains all positive pathogen species detected in the case within the corresponding time period, constituting a multinomial form of pathogen combination input. Subsequently, the association rule mining algorithm is executed on each batch of case sets to identify frequent itemsets where multiple pathogens co-occur in the same case. During the association rule mining process, to ensure that pathogen combinations have sufficient statistical significance, a minimum occurrence threshold for frequent itemset identification needs to be set. For example, the minimum support threshold is set to a fixed proportion of the total number of cases in the batch, such as 3%, and remains unchanged during the implementation phase. The identified frequent itemsets represent pathogen combinations with a high co-occurrence probability within the peak period. The joint support of these itemsets, i.e., the proportion of the combination in all cases, is calculated, and the pattern strength index of each combination is further calculated. The pattern strength index is obtained by comparing the ratio between the joint support of the combination and the independent support of each individual pathogen in the combination. For example, if the joint support of a combination significantly exceeds the independent support of a single pathogen, it indicates that the combination exhibits a significant co-occurrence characteristic. The screening threshold for the pattern intensity index is determined based on the actual pattern intensity index distribution, for example, set to 1.3. When the pattern intensity index of an itemset exceeds this fixed threshold, it is determined to be a pathogen combination with significant co-infection significance.

[0024] After completing the pattern intensity index screening, all pathogen co-occurrence combinations that meet the criteria are directly stored in the local co-infection pattern feature library. The feature library uses a structured set format, with each co-infection pattern including the types of pathogens that make up the pattern and the specific high-incidence time period identifier of their origin. This ensures that the feature library can correctly associate risk characteristics of different regions and time periods during subsequent matching. When constructing the feature library, if the same pathogen combination is found in multiple high-incidence time periods, it is treated as the same pattern entry and its frequency is recorded for future comprehensive analysis. To ensure the scalability of the feature library, a multi-segment storage method is used, where each high-incidence time period forms an independent co-infection pattern set. During the integration phase, all sets are aggregated to form a local co-infection pattern feature library covering the entire historical period. Each pattern record in the feature library is represented by explicit fields, including a list of pathogen combinations, a pattern intensity index value, and a high-incidence time period label.

[0025] In S3, the baseline of high-incidence periods in all regions is integrated to establish a three-dimensional collaborative matrix of regions, high-incidence periods and pathogens. The synchronous high-incidence region groups with different pathogen spectra are identified through spatiotemporal modal decomposition.

[0026] For each region, baseline data for identified high-incidence periods are extracted line by line to determine the actual time intervals corresponding to each high-incidence period. Historical pathogen detection data within these time intervals are aggregated, and the detection frequency of different pathogen types within each time period is statistically analyzed. The frequency value is used as the proportion of positive detections to all samples tested for that pathogen, ensuring comparability of detection frequencies across different time periods. During the statistical process, each pathogen is recorded individually to avoid the impact of simultaneous occurrence of certain pathogen combinations on its individual detection frequency. Using the above statistical methods, a set of individual pathogen frequencies for each region within each high-incidence period is formed. Subsequently, this frequency set is sorted by value from highest to lowest, and a set of dominant pathogens is determined according to the pre-defined dominant pathogen selection rules. Pathogens with detection frequencies exceeding a fixed threshold or ranking among the top in the frequency ranking are designated as dominant pathogens to ensure that the dominant pathogen set accurately reflects the representative pathogenic factor group within that time period. During annotation, the content of the dominant pathogen set is recorded in a structured manner, including pathogen name, detection frequency, and region and time period identifier. After forming a complete set of dominant pathogens, the pathogen spectrum corresponding to each region and each high-incidence period is constituted. The pathogen spectrum is represented by explicit fields, including a list of pathogens, their respective detection frequencies, and a dominant identifier.

[0027] The constructed pathogen spectrum is organized along three dimensions: region, peak incidence period, and pathogen type, forming a three-dimensional collaborative tensor. The region dimension represents different administrative areas, the peak incidence period dimension represents time windows within a region exhibiting the same peak incidence characteristics, and the pathogen type dimension includes all pathogen types involved in the analysis. Each element of the three-dimensional collaborative tensor is represented by a binary identifier: if a pathogen is identified as the dominant pathogen in its corresponding peak incidence period within a region, it is assigned a value of "1"; otherwise, it is assigned a value of "0". This format ensures the tensor's structure is clear, facilitating subsequent spatiotemporal analysis based on the distribution of dominant pathogens. After constructing the tensor, a spatiotemporal constraint clustering operation is performed using a clustering algorithm to identify initial clusters of regions that simultaneously satisfy temporal synchronicity and spatial correlation. The temporal synchronicity constraint requires that regions assigned to the same cluster must have significant overlap in their peak incidence periods on the time axis. The overlap ratio can be determined by setting specific thresholds, such as requiring the length of the overlapping interval between the peak incidence periods of two regions to be a fixed proportion of the shorter time period. Spatial connectivity constraints require that the clustered regions have direct boundaries or adjacency in geographic space, determined by administrative division information, such as shared boundaries or natural connections formed by specific transportation routes. When two regions simultaneously satisfy both constraints, they can be grouped into the same initial cluster. By scanning the structure of the three-dimensional cooperative tensor and classifying all regions according to the above constraint rules, initial clusters of synchronously high-incidence regions are ultimately formed, each consisting of multiple regions that satisfy the spatiotemporal constraints and recorded in the form of sets.

[0028] After obtaining the initial clusters of synchronously high-incidence areas, the pathogen spectrum difference between regions within each cluster is calculated. Specifically, for any two regions in a cluster, their respective dominant pathogen sets are extracted, and the differential pathogens between the sets are statistically analyzed. Differential pathogens refer to pathogens that exist only in the dominant set of one region and not in the set of the other region. Subsequently, the pathogen spectrum difference between the two regions is obtained by calculating the proportion of the number of differential pathogens to the union of the dominant pathogen sets of the two regions. For example, if the union of the two regions contains 6 pathogens, and 2 of them only appear in one region, the difference is the ratio of 2 to 6. Performing the above calculation on all pairs of regions within the same cluster yields a complete pairwise difference distribution. Subsequently, the average difference values ​​of all pairs within each cluster are averaged to obtain the average difference index of that cluster. The difference threshold is fixed before implementation, and the specific setting is based on the actual difference distribution, such as 0.3, 0.4, or other values. When the average difference of a cluster exceeds a set threshold, it indicates that there are significant differences in the dominant pathogen spectrum among the regions within the cluster. Such clusters are then labeled as groups of synchronously high-incidence regions with significant differences. The labeling method involves adding tags to the cluster structure to ensure complete identification. Finally, through the aforementioned screening process, multiple groups of synchronously high-incidence regions with significant differences are formed.

[0029] In step S4, real-time pathogen detection data from each region within the synchronous high-incidence area group is captured and matched with the corresponding local co-infection pattern feature library to generate a real-time co-infection risk list.

[0030] For each region designated as a high-incidence area, information on positive cases is continuously extracted in real-time from its pathogen detection records. The real-time extraction process follows a unified set of rules, including all records with positive results obtained within the current time window, regardless of whether the test originated from outpatient, emergency, inpatient, or centralized screening sites. Each record contains a clearly defined pathogen type field. To ensure a consistent pathogen set structure across regions, duplicate positive detections of the same case number within the same time window are deduplicated. For example, if the same pathogen is detected positive multiple times, the first record is retained, and duplicate records are deleted. After deduplication, all currently positive pathogen types in the region are organized into a set, forming the real-time pathogen type set for the current moment.

[0031] After obtaining the real-time pathogen sets for each region, these sets are used as input data and compared item by item with the local co-infection pattern feature library already constructed for the corresponding region. Each record in the feature library contains a clear combination of pathogen co-occurrences as a unit pattern, such as a combination pattern consisting of two, three, or more pathogens. During the comparison, for each co-infection pattern unit, it is directly checked whether any pathogen in the pattern appears in the current real-time pathogen set. When a pattern unit and the real-time set have at least one common pathogen, the pattern is determined to have characteristics that are partially hit by the real-time data, and the pattern is classified as a risk item. To fully characterize the potential infection structure of the pattern unit, this embodiment further annotates pathogens in the pattern that do not appear in the real-time pathogen set while recording risk items, defining them as permeable pathogens. This annotation method can clearly distinguish between pathogens that have already appeared in the region and pathogens that may infiltrate in subsequent stages. After completing the comparison and annotation of all co-infection pattern units, all eligible risk items are organized into a structured real-time co-infection risk list according to predetermined rules. Each record in the list includes the pattern pathogen combination, the current hit status, the list of permeable pathogens, and the high-incidence time period identifier of the pattern source.

[0032] In S5, the real-time risk lists of each region are cross-referenced and compared with the high-incidence area group. Combined with the intensity of inter-regional personnel flow, the imported co-infection pattern is deduced and the cross-regional risk penetration rate is calculated.

[0033] A real-time co-infection risk list was extracted from all regions within the synchronous high-incidence area group and organized into a structured manner according to region. Using existing population movement relationships between regions as the pairing basis, pathogen combination pairing analysis was performed on each pair of regions with actual population movement directions. For each source region and destination region pairing, firstly, all pathogens marked as permeable in the destination region's real-time risk list were extracted and organized into a permeable pathogen set. Next, all existing pathogen co-occurrence combinations in the source region's real-time risk list were extracted; each combination consists of several pathogens already present in the source region. After extracting both sets, the permeable pathogens in the destination region were sequentially combined with the pathogen co-occurrence combinations in the source region through a one-to-one comparison method to construct potential pathogen co-occurrence combinations that might be activated in the destination region. The combination method employed a structural splicing approach, merging the pathogen combinations in the source region with the permeable pathogens in the destination region into a new combination, while filtering out duplicate pathogens to ensure that each pathogen appears only once in each combination and that a corresponding pathogen co-occurrence combination can be found in the corresponding local co-infection pattern feature library. Through the above operations, a series of potential risk pathogen combinations for the destination region can be generated after expansion of pathogen combinations from the source region. To ensure the stability of the combination structure, each combination record is stored as a structured entry and appended with a region-to-region relationship identifier. All of the above potential combinations are uniformly regarded as imported co-infection patterns, representing pathogen co-occurrence structures that may form in the destination region under conditions of population movement. Finally, these imported co-infection patterns are output as a set.

[0034] After generating the imported co-infection pattern, the cross-regional risk penetration rate is calculated for each item's potential risk pathogen combination to quantify the likelihood of cross-regional transmission. The calculation first obtains the intensity of population flow from the source region to the destination region. This intensity is provided by inter-regional population migration statistics, traffic flow monitoring, or administrative records and recorded as numerical fields in a fixed format. Then, the frequency of permeable pathogens appearing in all risk items in the destination region's real-time risk list is statistically analyzed. This frequency is obtained using a fixed calculation method, such as representing the proportion of the number of times the pathogen appears in all risk items to the total number of risk items. To ensure consistency across all indicators, the population flow intensity and pathogen frequency are dimensionless, allowing for direct multiplication. Finally, the cross-regional risk penetration rate corresponding to the pathogen combination is obtained by calculating the dimensionless product of the population flow intensity and the permeable pathogen frequency. To ensure bidirectional results, this step uses the same method for reverse analysis, pairing and expanding the permeable pathogen co-occurrence combinations in the source region's risk list with existing pathogen co-occurrence combinations in the destination region's risk list in another direction, and calculating the reverse cross-regional risk penetration rate in the same way. Reverse penetration rate calculation follows a completely consistent process, including pathogen combination expansion, permeable pathogen frequency statistics, acquisition of population flow intensity, and dimensionless product calculation. All forward and reverse penetration rates are recorded as independent entries, ensuring that each pair of regions has bidirectional risk measurement.

[0035] In S6, a visualized dynamic risk heat map is established based on the imported co-infection pattern and the corresponding cross-regional risk penetration rate.

[0036] All potential risk pathogen co-occurrence combinations and their corresponding cross-regional risk penetration rates are comprehensively collected according to region pairs. Since this data includes both the risk penetration rate from the source region to the destination region and the reverse penetration rate, it needs to be structured and stored in a bidirectional manner during the data processing. For example, two independent data entries are recorded in the form of "Region A → Region B" and "Region B → Region A," and the corresponding potential risk pathogen combinations are associated with their calculated risk penetration rates by field. After collection, a set of bidirectional imported co-infection patterns covering all region pairs is formed. This set contains all pathogen combination structures that may be affected by cross-regional transmission, providing a complete data foundation for subsequent risk mapping. Subsequently, a dynamic risk heatmap is constructed in the visualization layer. The heatmap uses geographical regions as the basic unit and the risk penetration rate between regions as the numerical driving source. Layer rendering is achieved by mapping the penetration rate of each region pair to a geographical coordinate region. The penetration rate values ​​are mapped to corresponding color values ​​from low to high at a fixed ratio. During heatmap generation, each region is checked individually. When a region receives cross-regional risk penetration rates from any source region as the destination region, its penetration rate value is written into the heatmap node field of that region. If a region has multiple source regions, all penetration rates are recorded in parallel. After the overall rendering is completed, a dynamically updated risk heatmap is obtained, which is used for visual pre-annotation of respiratory infection risks in children.

[0037] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0038] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0039] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0040] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0041] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0042] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0043] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 this application. 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.

[0044] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0045] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the risk of respiratory infections in children based on multi-source medical data, characterized in that, Includes the following steps: S1. Extract historical electronic medical records of children's respiratory infections and pathogen detection data from local medical institutions in various regions to construct a baseline for the high incidence period of children's respiratory infections in the local area; S2. Perform time-segment analysis on historical pathogen-positive cases in various regions to identify pathogen co-occurrence combinations during each peak period of respiratory infection and construct a local co-infection pattern feature library. S3. Integrate the baseline of high-incidence periods in all regions, establish a three-dimensional collaborative matrix of regions, high-incidence periods and pathogens, and identify synchronous high-incidence regions with different pathogen spectra through spatiotemporal mode decomposition; S4. Capture real-time pathogen detection data from each region within the synchronous high-incidence area group, match it with the corresponding local co-infection pattern feature library, and generate a real-time co-infection risk list. S5. Using high-incidence areas as units, cross-compare the real-time risk lists of each region, combine the intensity of inter-regional personnel flow, deduce the imported co-infection pattern, and calculate the cross-regional risk penetration rate. S6. Based on the imported co-infection pattern and the corresponding cross-regional risk penetration rate, establish a visualized dynamic risk heat map.

2. The method for predicting the risk of respiratory infections in children based on multi-source medical data according to claim 1, characterized in that, In step S1, extracting historical electronic medical records of children's respiratory infections and pathogen detection data from local medical institutions in each region to construct a baseline for high-incidence periods of local children's respiratory infections specifically includes: The pathogen detection data includes target pathogen types, detection frequency, and detection result data within a preset range; Extract pediatric respiratory infection data from historical electronic medical records in time series, calculate the moving average ratio of the number of cases in each time period to the number of cases in adjacent time periods, and use it as an indicator of local activity. The local activity index is combined with the historical average number of cases in the same period to establish a comprehensive time-series feature vector; Density clustering algorithm was used to analyze the distribution of the comprehensive temporal feature vector on the time axis, identify continuous time periods with feature vector density higher than the average level, and integrate them into a baseline of high-incidence periods of respiratory infections in local children.

3. The method for predicting the risk of respiratory infections in children based on multi-source medical data according to claim 1, characterized in that, In S2, time-segment analysis is performed on historical pathogen-positive cases in various regions to identify pathogen co-occurrence combinations during each peak period of respiratory infection, and to construct a local co-infection pattern feature library, specifically including: Extract the high-incidence time periods from the baseline of high-incidence periods in each region, aggregate all historical pathogen detection data with positive test results within each high-incidence time period, and mark the corresponding cases as positive cases; Using association rule mining algorithms, we analyze frequent itemsets in each batch of positive cases where multiple pathogens co-occur, and calculate the joint support of each itemset. The pattern strength index is calculated based on the ratio of joint support to the independent support of each pathogen in the itemset, and pathogen co-occurrence combinations with pattern strength indices exceeding a preset threshold are screened. Pathogen co-occurrence combinations are stored in a feature library, a set of co-infection patterns for high-incidence time periods is established, and all sets of high-incidence time periods are aggregated to form a local co-infection pattern feature library.

4. The method for predicting the risk of respiratory infections in children based on multi-source medical data according to claim 1, characterized in that, In S3, the baseline of high-incidence periods in all regions is integrated to establish a three-dimensional collaborative matrix of regions, high-incidence periods, and pathogens. Specifically, synchronous high-incidence region groups with differences in pathogen spectra are identified through spatiotemporal mode decomposition, including: Extract the high-incidence time periods from the baseline of high-incidence periods in each region, and the detection frequency of different types of single pathogens in the corresponding historical pathogen detection data, construct a pathogen spectrum and label the dominant pathogen set; The baseline of high-incidence periods in all regions is collected, and a three-dimensional collaborative tensor with region, high-incidence period and pathogen type as the axis is constructed. Spatiotemporal constraint clustering is performed on the three-dimensional collaborative tensor to extract the initial cluster of synchronous high-incidence regions. The spatiotemporal constraints include the synchronicity constraint of high-incidence period and the spatial connectivity constraint in the region dimension. Calculate the pathogen spectrum difference between any two regions within the initial cluster of regions based on the dominant pathogen set, where the difference is the proportion of the number of asymmetrically different pathogens in the dominant pathogen sets of the two regions to the total number of pathogens in the union set. Clusters in which the mean difference in pathogen spectrum between all regions within the initial cluster of the selected region is greater than a set value are labeled as synchronously high-incidence regions with significant differences.

5. The method for predicting the risk of respiratory infections in children based on multi-source medical data according to claim 4, characterized in that, The value of each element in the three-dimensional cooperative tensor is a binary identifier, which indicates whether the pathogen is listed as the dominant pathogen in its pathogen spectrum characteristics in the corresponding region during the corresponding high-incidence period.

6. The method for predicting the risk of respiratory infections in children based on multi-source medical data according to claim 1, characterized in that, In step S4, real-time pathogen detection data from various regions within the synchronous high-incidence area group is captured and matched with the corresponding local co-infection pattern feature library to generate a real-time co-infection risk list, specifically including: Real-time extraction of pathogen-positive case data for each region within the synchronous high-incidence area group, generating a current deduplicated real-time pathogen type set for each region; The real-time pathogen set is compared with the pathogen co-occurrence combination of each co-infection pattern unit in the local co-infection pattern feature library of the corresponding region. The pathogen co-occurrence combination with intersection is used as a risk item, and the pathogens that are not hit in the pathogen co-occurrence combination are marked as permeable pathogens. All selected risk items are aggregated to generate a real-time co-infection risk list for each region.

7. The method for predicting the risk of respiratory infections in children based on multi-source medical data according to claim 1, characterized in that, In S5, taking high-incidence areas as units, the real-time risk lists of each region are cross-referenced, and the intensity of inter-regional population movement is combined to deduce imported co-infection patterns and calculate cross-regional risk penetration rates. Specifically, this includes: Extract the real-time co-infection risk list for each region within the high-incidence region group. For each pair of regions with personnel flow within the group, pair the pathogens marked as permeable in the risk list of the destination region with the existing pathogen co-occurrence combinations in the risk list of the source region to generate a potential risk pathogen co-occurrence combination of the destination region activated by adding the permeable pathogens of the destination region to the original pathogen co-occurrence combination in the risk list of the source region. Co-occurrence combinations of potentially risky pathogens are considered as imported co-infection patterns; The cross-regional risk penetration rate of potential risk pathogen co-occurrence combinations in the destination area is calculated by multiplying the intensity of personnel flow from the source area to the destination area by the dimensionless product of the frequency of occurrence of the permeable pathogen in all risk items of the risk list in the destination area. Using the same method, pathogens marked as permeable in the source region risk list are reverse-paired with existing pathogen co-occurrence combinations in the destination region risk list, and the cross-regional risk penetration rate is calculated.

8. The method for predicting the risk of respiratory infections in children based on multi-source medical data according to claim 1, characterized in that, In S6, establishing a visualized dynamic risk heatmap based on imported co-infection patterns and corresponding cross-regional risk penetration rates specifically includes: By aggregating potential risk pathogen co-occurrence combinations generated from all directions and their corresponding cross-regional risk penetration rates, a two-way imported co-infection pattern is formed and updated in real time, establishing a visualized dynamic risk heat map of respiratory infections in children.

Images