Method, device and computer program product for managing a population health profile

Abstract

This application proposes a management method, management device, and computer program product for population health profiling, relating to the field of population profiling technology. The management method includes: acquiring multi-source heterogeneous data within a preset collection area, including socioeconomic dimensions, health behavior dimensions, environmental exposure dimensions, and medical service dimensions; constructing feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, and generating a subgroup health profile corresponding to each feature subgroup; extracting historical time-series data from the subgroup health profiles and inputting the historical time-series data into a preset early warning model to generate corresponding health risk trends and graded early warning signals; and matching graded management strategies corresponding to the health risk trends and graded early warning signals through a preset strategy knowledge base. This application can deeply integrate multi-dimensional data, accurately characterize population health features, predict the evolution of health risks in real time, and achieve intelligent graded management.

Description

Technical Field

[0001] This application relates to the field of population profiling technology, and in particular to a method, device and computer program product for managing population health profiles. Background Technology

[0002] In the practice of chronic disease prevention and health management at the county level, we face multiple deep-seated challenges. There are significant differences among the populations of different townships and communities within a county in terms of socioeconomic background, daily behavior patterns, environmental exposure conditions, and access to medical services. This high heterogeneity leads to an uneven distribution of health risk factors, and the traditional management model, which adopts uniform intervention methods, cannot adapt to the personalized health needs of different groups. Summary of the Invention

[0003] The main objective of this invention is to provide a management method for population health profiles, which aims to deeply integrate multi-dimensional data, accurately depict population health characteristics, predict the evolution of health risks in real time, and achieve intelligent hierarchical management.

[0004] To achieve the above objectives, the present invention provides a method for managing population health profiles, the method comprising: Acquire multi-source heterogeneous data within a preset collection area, the multi-source heterogeneous data including socioeconomic dimensions, health behavior dimensions, environmental exposure dimensions, and medical service dimensions; Construct feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, and generate a subgroup health profile corresponding to each feature subgroup; Historical time-series data are extracted from the subgroup health profiles and input into a preset early warning model to generate corresponding health risk trends and graded early warning signals. A hierarchical management strategy is matched with the health risk trend and the hierarchical early warning signal by a preset strategy knowledge base.

[0005] Optionally, the step of constructing feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, and generating a subgroup health profile corresponding to each feature subgroup, includes: Construct a comprehensive quantitative indicator system covering every data point in the multi-source heterogeneous data; The comprehensive quantitative index system is subjected to feature dimensionality reduction processing by principal component analysis or factor analysis to extract multiple potential core common factors that can characterize the determinants of population health. Based on clustering algorithms and multiple potential core common factors, the population within the preset collection area is stratified and divided into multiple feature subgroups that are internally homogeneous but different from each other. Generate a multi-dimensional subgroup health profile corresponding to each of the aforementioned feature subgroups. The multi-dimensional subgroup health profile includes the distribution of key feature indicators of the corresponding subgroup, a list of major health problems, analysis of potential risk drivers, and preliminary intervention direction suggestions.

[0006] Optionally, the population within the preset collection area is stratified based on a clustering algorithm and multiple potential core common factors to generate multiple feature subgroups that are internally homogeneous but differentiated from each other, including: A prototype clustering algorithm capable of simultaneously processing numerical continuous variables and categorical discrete variables is adopted as the basic partitioning algorithm; Based on the data characteristics of the potential core common factors, a hybrid dissimilarity measurement function suitable for the prototype clustering algorithm is constructed; The prototype clustering algorithm was iteratively verified to determine the optimal number of feature subgroups by using a combination of silhouette coefficient and Davidson-Bolding index as evaluation metrics. The partitioning results of the prototype clustering algorithm are optimized by using a density-based noise spatial clustering algorithm to identify and exclude discrete noise points, thereby generating the feature subgroup.

[0007] Optionally, the step of extracting historical time-series data from the subgroup health profile and inputting the historical time-series data into a preset early warning model to generate corresponding health risk trends and graded early warning signals includes: Extract the first observation of key health outcome indicators and the second observation of core risk factor exposure levels at consecutive time points from the subgroup health profile of each subgroup; Construct a multivariate time series dataset corresponding to the first observation and the second observation; For each data point in the multivariate time series dataset, sliding window processing, seasonal differencing, and stationarity tests are performed to generate corresponding historical time series data. The historical time series data is input into the preset early warning model, and the health risk trend within a specific future time span is output. The health risk trend is compared with the preset multi-level risk thresholds to generate the graded early warning signal.

[0008] Optionally, the extraction of the first observation of key health outcome indicators and the second observation of core risk factor exposure levels at consecutive time points in the subgroup health profile of each subgroup includes: From the list of major health problems in the subgroup health profile, the prevalence of hypertension, the incidence of diabetes, and the mortality rate of cardiovascular and cerebrovascular diseases were extracted as the first observations of key health outcome indicators. From the distribution of key characteristic indicators of the subgroup health profile, the smoking rate of the adult population, the average body mass index, and the participation rate in moderate to high intensity physical exercise were extracted as the second observation of the exposure level of core risk factors.

[0009] Optionally, the method for managing the population health profile further includes: Construct a neural network infrastructure with a long short-term memory network or gated recurrent unit as its core, the neural network infrastructure being used to capture long-term dependencies in time series; An attention mechanism is incorporated into the neural network infrastructure. This attention mechanism is used to dynamically evaluate the importance of the hidden state at different historical time points to the current prediction task during the forward computation process and generate corresponding weight allocation results. The weighted fusion of all hidden states output by the neural network infrastructure is performed using the weighted allocation results. A multi-objective loss function consisting of mean squared error loss function and mean absolute percentage error loss function is defined. An adaptive moment estimation optimization algorithm is used to train the entire network parameters, including the attention mechanism, end-to-end. During model training, an independent validation dataset is used to evaluate the model's prediction accuracy, and models with a root mean square error in the test set below a preset threshold are saved as the preset early warning models.

[0010] Optionally, the step of matching the hierarchical management strategy corresponding to the health risk trend and the hierarchical early warning signal through a preset strategy knowledge base includes: Generate a strategy query request corresponding to the health risk trend and the graded early warning signal; The strategy query request is input into the preset strategy knowledge base for matching and retrieval to determine one or more most suitable combinations of hierarchical management strategies; Resource consumption simulation and preliminary feasibility assessment are performed on the combination of hierarchical management strategies, and the hierarchical management strategies are output.

[0011] Optionally, the step of performing resource consumption simulation and preliminary feasibility assessment on the combination of hierarchical management strategies, and outputting the hierarchical management strategy, includes: Acquire real-time distribution data and available capacity data of medical and health resources within the preset collection area; The resource consumption simulation model is used to quantitatively estimate the various resources required for each measure in the hierarchical management strategy combination, and the corresponding resource demand estimation results are generated. The resource demand estimation results are compared and analyzed with real-time distributed data and available capacity data to identify potential resource bottlenecks and conflicts, and to determine the comparison and analysis results. Based on the comparison and analysis results, the hierarchical management strategy combination is adjusted and optimized to generate the hierarchical management strategy.

[0012] In addition, to achieve the above objectives, the present invention also provides a management device, the management device comprising: a memory, a processor, and a management program for a population health profile stored in the memory and executable on the processor, the management program for the population health profile being configured to implement the population health profile management method as described above.

[0013] In addition, to achieve the above objectives, the present invention also provides a computer program product, including the management device described above.

[0014] This invention acquires multi-source heterogeneous data within a preset collection area, including socioeconomic, health behavior, environmental exposure, and medical service dimensions. It then constructs feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, generates a subgroup health profile for each feature subgroup, extracts historical time-series data from the subgroup health profiles, and inputs this historical time-series data into a preset early warning model to generate corresponding health risk trends and tiered early warning signals. Finally, a preset strategy knowledge base is used to match the tiered management strategy corresponding to the health risk trends and tiered early warning signals. Thus, by integrating multi-source heterogeneous data to construct feature subgroups and generate health profiles, and combining time-series data analysis and early warning models to achieve dynamic risk prediction, and ultimately matching tiered management strategies, this invention effectively solves the problems of data dispersion, high population heterogeneity, delayed risk prediction, and extensive resource allocation in county-level chronic disease prevention and control. It can deeply integrate multi-dimensional data, accurately characterize population health characteristics, predict the evolution of health risks in real time, and achieve intelligent tiered management. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0016] To more clearly illustrate the technical solutions in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a population health profile management method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a population health profile management method according to another embodiment of the present invention; Figure 3 A schematic diagram of a population health profile management method according to another embodiment of the present invention; Figure 4 This is a schematic diagram of a population health profile management method according to another embodiment of the present invention; Figure 5 A schematic diagram of a population health profile management method according to another embodiment of the present invention; Figure 6 This is a schematic diagram of a population health profile management method according to another embodiment of the present invention; Figure 7 A schematic diagram of a population health profile management method according to another embodiment of the present invention; Figure 8 This is a schematic diagram of a population health profile management method according to another embodiment of the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] 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, not all embodiments. Well-known modules, units, and their connections, links, communications, or operations are not shown or described in detail. Furthermore, the described features, architectures, or functions can be combined in any way in one or more embodiments. Those skilled in the art should understand that the various embodiments described below are only for illustrative purposes and not for limiting the scope of protection of the present invention. It is also readily understood that the modules, units, or processing methods in the various embodiments described herein and shown in the accompanying drawings can be combined and designed in various different configurations. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The definitions of various terms or methods used in the following embodiments are, except where logically impossible, generally defined as broad concepts that can be implemented under the premise of the content disclosed in the embodiments. Under this understanding, all specific subordinate limitations of the terms or methods should be considered as part of the invention and should not be narrowly interpreted or biased simply because the specification does not disclose such a specific limitation. Similarly, provided that it is logically feasible, the order of the steps in the method is flexible and varied, and all specific subordinate limitations in the broad concepts of various terms or methods fall within the scope of protection of this invention.

[0021] This invention belongs to the interdisciplinary fields of public health informatics, health big data analysis, the application of artificial intelligence in the medical and health field, chronic disease management, regional health governance, and data mining and machine learning. Specifically, it relates to a system and method that can integrate multi-dimensional data from a county, construct population health profiles through intelligent clustering, use time-series prediction models for risk warning, and realize dynamic hierarchical management and adaptive optimization of intervention strategies based on this.

[0022] However, counties generally face the following challenges in chronic disease prevention and health management: 1. Population Heterogeneity and Diverse Needs: Within a county, significant differences exist among populations in different townships and villages in terms of socioeconomic level (income, education, occupation), lifestyle habits (diet, exercise, smoking and alcohol consumption), environmental exposure (air, water quality, green space), health literacy, and accessibility and utilization of medical services. This heterogeneity leads to uneven distribution of chronic disease risk factors, and different subgroups have varying health needs and intervention priorities, making traditional "one-size-fits-all" management models ineffective.

[0023] 2. Data Silos and Dimensional Limitations: The data required for health management decisions is often scattered across multiple departments, including public health information systems, hospital HIS / EMR, health checkup centers, civil affairs, statistics, and environmental protection. Data standards vary, making integration difficult. Furthermore, traditional analyses often focus on clinical data, failing to adequately consider broader socioeconomic, behavioral, and environmental determinants of health (SDOH), thus failing to comprehensively reflect the overall health of the population.

[0024] 3. Delayed risk identification and passive response: Existing health management systems rely heavily on analysis of reported diseases or annual physical examination data, lacking dynamic monitoring of trends in population health status and proactive prediction and early warning capabilities for future risks (such as the onset, exacerbation, and complications of specific chronic diseases). This results in intervention measures often lagging behind the occurrence of risks.

[0025] 4. Inefficient and extensive resource allocation: Due to the lack of detailed population profiles and accurate risk stratification, limited public health and medical resources are difficult to accurately target the high-risk subgroups and key links that need the most help, resulting in low cost-effectiveness of intervention measures and limited overall prevention and control effectiveness.

[0026] 5. Lack of agility and data-driven approach in strategy adjustments: The formulation and adjustment of chronic disease prevention and control strategies often rely on long-term statistical reports and expert experience, lacking a dynamic optimization mechanism based on real-time data feedback. This makes it difficult to quickly adapt to changes in population health status, feedback on intervention effects, and emerging health threats.

[0027] Therefore, there is an urgent need for an innovative solution that can integrate multidimensional data, accurately depict the health profile of the population, dynamically predict risks, and achieve intelligent, personalized, and adaptive hierarchical management to improve the scientificity and effectiveness of chronic disease prevention and control in counties.

[0028] This invention aims to provide a county-level population health profile construction and dynamic hierarchical closed-loop management system and method based on multi-dimensional indicator fusion, time-series prediction, and adaptive strategy optimization. Its purpose is: To achieve systematic integration and scientific evaluation of multi-dimensional indicators such as socioeconomic status, health behavior, environmental exposure, and medical services of the population in the county.

[0029] Through advanced clustering algorithms, we can intelligently identify homogeneous subgroups within a county that have different health risk characteristics and intervention needs, and construct accurate and interpretable population health profiles.

[0030] By using time-series prediction models (such as LSTM combined with attention mechanisms), we can dynamically monitor the changing trends of health indicators in various subgroups and predict their future risk of chronic disease occurrence or exacerbation.

[0031] Based on population profiles and risk warning results, we can intelligently recommend or generate differentiated and personalized hierarchical management strategies and resource allocation schemes for different subgroups.

[0032] Establish a closed-loop management system that can adaptively adjust grading standards, predictive models, and intervention strategies based on feedback on intervention effects and dynamic changes in population health.

[0033] Ultimately, this will improve the scientific nature of county-level public health decision-making, the precision of chronic disease management, the initiative in risk warning, and the efficiency of resource allocation.

[0034] The main solution of this application embodiment is as follows: by acquiring multi-source heterogeneous data within a preset collection area, wherein the multi-source heterogeneous data includes socioeconomic dimensions, health behavior dimensions, environmental exposure dimensions, and medical service dimensions, feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data are constructed, and a subgroup health profile corresponding to each feature subgroup is generated. Then, historical time-series data is extracted from the subgroup health profiles, and the historical time-series data is input into a preset early warning model to generate corresponding health risk trends and graded early warning signals. Finally, a graded management strategy corresponding to the health risk trends and graded early warning signals is matched through a preset strategy knowledge base.

[0035] To achieve the above objectives, the method proposed in this invention mainly includes the following core steps: 1. Module for Dynamic Collection and Integrated Governance of Multidimensional Health-Related Indicators at the County Level: Automatic connection and aggregation of multi-source heterogeneous data: Design and implement standardized interfaces with various data sources within the county, including but not limited to: public health information systems (such as chronic disease registration, maternal and child health care, immunization programs), hospital HIS / EMR systems, regional health record platforms, business systems of township health centers / community health service centers, medical insurance settlement data, civil affairs department data (such as population structure, low-income population), statistics bureau data (such as income, education, employment), environmental protection department data (such as air quality, water quality monitoring), as well as behavioral lifestyle data and health literacy data collected supplementarily through mobile health apps, wearable devices, customized surveys, etc.

[0036] Data cleaning, standardization, and spatiotemporal alignment: The collected raw data is deduplicated, denoised, missing value processed, and format converted. Using unified medical terminology standards (such as ICD, SNOMED CT, LOINC) and geocoding standards, the data is semantically standardized and spatiotemporally aligned to construct a dynamic health-related database based on individuals or families and linked to the smallest administrative division (such as village / community).

[0037] Privacy and security: In the process of data collection, storage, processing and sharing, we strictly abide by data security and privacy protection regulations and adopt technical means such as anonymization, pseudonymization, data encryption and access control.

[0038] 2. A module for constructing intelligent profiles of county-level population subgroups based on the integration of multi-dimensional indicators including socio-economic, health, behavior, and environment (Innovation Point 1): Construction and Feature Engineering of Multidimensional Indicator System: Systematically sort out and construct a core indicator system covering four dimensions: socio-economic (income level, education level, occupation type, housing conditions, social security participation), health status (morbidity, abnormal rate of physical examination indicators, frequency / cost of medical service use, vital signs), behavior and lifestyle (smoking rate, drinking rate, rationality of dietary structure, frequency of physical exercise, health literacy level), and environmental exposure (air quality index AQI, water quality compliance rate, green coverage rate, noise level, accessibility of public service facilities).

[0039] Factor analysis and feature dimensionality reduction: Principal component analysis (PCA), factor analysis (EFA), or nonlinear dimensionality reduction methods (such as t-SNE, UMAP) are used to process high-dimensional indicators, extract potential core factor / feature combinations that can effectively reflect the health status of the population and its complex influencing factors, reduce data redundancy, and improve the efficiency and interpretability of subsequent clustering.

[0040] Multimodal / mixed data hierarchical clustering and subgroup identification: For indicators containing multiple data types, including continuous, categorical, and ordered data, advanced clustering algorithms are employed, such as: Improved hierarchical clustering algorithms (such as AGNES and DIANA) can be combined with appropriate distance metrics and connection methods.

[0041] Clustering algorithms for mixed-type data (such as K-Prototypes) can handle both numerical and categorical attributes simultaneously.

[0042] Density- or graph-based clustering algorithms (such as DBSCAN, Spectral Clustering) are used to discover clusters of arbitrary shapes.

[0043] The goal is to intelligently divide county-level populations or different geographical units (such as townships / villages) into several statistically significant subgroups with high internal homogeneity based on the "socioeconomic-health-behavioral-environment" characteristic spectrum (e.g., "high-income-good health behavior-low environmental risk subgroup", "low-income-multiple unhealthy lifestyle habits-high environmental exposure risk subgroup", etc.).

[0044] Automatic generation and visualization of subgroup health profiles: Multi-dimensional health profiles are automatically generated for each identified subgroup. The profile content includes: Key feature indicators of a subpopulation: Highlight the significant feature values ​​or distributions of the subpopulation across various dimensions.

[0045] Overview of major health problems and risks: This section summarizes the prevalence of major chronic diseases (such as hypertension and diabetes) and the prevalence of high-risk factors in this subgroup.

[0046] Analysis of potential driving factors: Preliminary findings reveal key socioeconomic, behavioral, or environmental factors influencing the health status of this subgroup.

[0047] Intervention needs and directions: Based on profile characteristics, preliminary suggestions are made regarding possible intervention priorities and directions.

[0048] Visual presentation: The profiles of subgroups are presented intuitively through radar charts, bar charts, heat maps, word clouds, and GIS map overlays.

[0049] 3. A dynamic monitoring and hierarchical early warning model for subgroup health risks based on LSTM and attention mechanisms (Innovation Point Two): Subgroup-level time-series data construction and feature engineering: For each profile subgroup, time-series data of key health outcome indicators (such as incidence / prevalence of specific chronic diseases, mortality, complication rate, and growth rate of medical expenses), exposure levels of major risk factors (such as changes in smoking rate and average BMI), and coverage rates of intervention measures (such as standardized management rate of hypertension and glycemic control rate of diabetic patients) are extracted from historical data. Time-series feature engineering is then performed, such as window sliding, differencing, and seasonal decomposition.

[0050] LSTM (or GRU) network construction and training: Construct a Long Short-Term Memory (LSTM) network or Gated Recurrent Unit (GRU) model, input historical time-series data of each subgroup, and learn its complex nonlinear temporal dependencies and dynamic change patterns. The model aims to predict the trends of key health outcome indicators for this subgroup over a future period (e.g., the next month, quarter, or year).

[0051] Attention Mechanism Integration: Introducing attention mechanisms (such as Bahdanau Attention, Luong Attention, and Self-Attention) into LSTM / GRU models. Attention mechanisms enable the model to automatically identify and assign higher weights to historical time points or influencing factors that contribute more to the current prediction result, thereby improving prediction accuracy and enhancing model interpretability (e.g., analyzing which historical factors or events have the greatest impact on future risk changes).

[0052] Dynamic risk assessment and multi-level early warning: Based on the output of the prediction model (such as predicted chronic disease incidence and probability of worsening trend), combined with preset risk thresholds (which can be set according to national / regional standards or historical baselines and can be dynamically adjusted) and the rate of change, the system performs dynamic, multi-level risk assessment and classification for each subgroup (e.g., extremely high risk, high risk, intermediate risk, low risk, and improving trend). When the predicted risk exceeds the early warning threshold or shows a significant worsening trend, the system automatically triggers an early warning and notifies relevant management personnel.

[0053] 4. Subgroup profiling and risk warning-driven adaptive hierarchical management strategy generation and optimization engine (Innovation Point 3): Construction of a Tiered Management Strategy Knowledge Base: Establish a structured and scalable knowledge base for chronic disease prevention and health management strategies. This base includes intervention packages targeting different health problems, risk levels, and subgroup characteristics. Package content may include: health education materials and activities, behavior modification techniques (such as smoking cessation interventions, dietary guidance, and exercise prescriptions), early screening programs (such as specific cancer screenings and hypertension / diabetes screenings), clinical management pathways and guidelines, community support services (such as patient support groups and rehabilitation resource links), and policy intervention recommendations (such as environmental improvement and health promotion policies). Association rules will be established between strategies and subgroup characteristics and risk levels.

[0054] Intelligent strategy recommendation and customized solution generation: When a subgroup is identified or its risk level changes, the system automatically retrieves, matches, and recommends one or more of the most suitable, differentiated hierarchical management strategy combinations from the strategy knowledge base, based on the subgroup's detailed profile characteristics (such as major health problems, key driving factors, and resource accessibility) and the current risk warning level. It supports manual adjustment and customization of the recommended strategies to form targeted intervention plans.

[0055] Resource optimization and allocation suggestions: Based on the distribution and accessibility data of existing medical and health resources (human resources, equipment, beds, medicines, and funding) in the county, the system can provide preliminary resource optimization and allocation suggestions for recommended strategy combinations. For example, which areas should prioritize screening resources, which populations need enhanced health education, and which primary healthcare institutions need technical support or personnel training.

[0056] Dynamic monitoring and evaluation of intervention effectiveness: Continuously track changes in key health indicators, behavioral indicators, and service utilization indicators for each target subgroup after the implementation of the tiered management strategy. Employ appropriate statistical methods (such as propensity score matching and difference-in-differences, DID) to evaluate the actual effectiveness of the intervention.

[0057] Closed-loop feedback and adaptive strategy adjustment: The evaluation results of the intervention effect are fed back to the system. If the effect does not meet expectations or new risk trends emerge, the system can: Adjusting population subgroups: Rerun the clustering algorithm to see if the subgroup definitions or numbers need to be updated.

[0058] Optimize prediction models: Retrain or fine-tune prediction models such as LSTM using new data to improve prediction accuracy.

[0059] Update the strategy knowledge base: Adjust the effectiveness rating and applicable conditions of the measures in the strategy base based on practical results, or introduce new effective intervention measures.

[0060] Adjust risk classification thresholds or early warning rules.

[0061] This closed-loop feedback mechanism enables continuous optimization and adaptive adjustment of management strategies.

[0062] In this embodiment, for ease of description, the following description will focus on the management device as the executing entity.

[0063] This application provides a solution that integrates multi-source heterogeneous data to construct characteristic subgroups and generate health profiles. It combines time-series data analysis and early warning models to achieve dynamic risk prediction and finally matches hierarchical management strategies. This effectively solves the problems of data dispersion, high population heterogeneity, lagging risk prediction, and extensive resource allocation in county-level chronic disease prevention and control. It can deeply integrate multi-dimensional data, accurately depict population health characteristics, predict the evolution of health risks in real time, and realize intelligent hierarchical management.

[0064] By integrating and innovating the above-mentioned technical solutions, this invention can bring the following significant technical effects and advantages: 1. Accurate population health profile construction: Innovatively integrating multi-dimensional indicators such as socio-economic, health, behavior, and environment, and using intelligent clustering methods, it can scientifically and objectively identify subgroups with different health characteristics and needs within the county, laying a solid foundation for refined management.

[0065] 2. Proactive dynamic early warning of health risks: The system pioneered the use of LSTM combined with deep learning models such as attention mechanisms to predict subgroup health risks over time, realizing the transformation from passive response to proactive prevention, improving the lead time and accuracy of risk identification, and enhancing the interpretability of the model.

[0066] 3. Intelligent adaptive hierarchical management strategy: A strategy generation and optimization engine driven by both profiling and risk has been constructed, which can provide differentiated and personalized intervention solutions for different subgroups. Through a closed-loop feedback mechanism, the strategy can continuously learn and adaptively adjust, thereby improving the pertinence and effectiveness of the intervention.

[0067] 4. Enhance the scientific nature of public health decision-making and the efficiency of resource allocation: Provide strong data-driven decision support for county-level health management departments, help them accurately identify priority groups and key issues, optimize the allocation of public health resources, and shift from "broad-based" to "precision-based" approaches.

[0068] 5. Promoting health equity and equal access to services: It helps identify and focus on vulnerable populations and resource-scarce areas, narrowing health disparities through targeted interventions and promoting equal access to basic public health services.

[0069] 6. Optimize the entire process and overall effectiveness of chronic disease management: Form a dynamic closed-loop management model of "data collection - profile construction - risk prediction - hierarchical management - effect evaluation - strategy optimization" to comprehensively improve the county's ability and level of integrated prevention and control of chronic diseases.

[0070] Strong scalability and promotional value: The system design has good modularity and configurability, which can be easily adjusted and deployed according to the actual conditions of different counties, and has broad prospects for promotion and application.

[0071] in: County-level health management departments (health commissions / bureaus): are responsible for formulating regional chronic disease prevention and control plans, accurately allocating public health resources (such as screening quotas for breast and cervical cancer, and funding for hypertension / diabetes management projects), dynamically monitoring the effectiveness of policy implementation, evaluating the effectiveness of intervention measures, and optimizing health promotion strategies.

[0072] County-level medical consortia / close-knit county-level medical alliances: serving as the core information platform for their refined health management of contracted residents within their jurisdiction, they guide subordinate medical institutions (township health centers, community health service centers, and village clinics) to carry out targeted health education, high-risk population screening, standardized management of chronic diseases, and follow-up services, thereby promoting hierarchical diagnosis and treatment and the integration of prevention and treatment.

[0073] Centers for Disease Control and Prevention (CDC): Used for conducting epidemiological surveillance of regional chronic diseases and their risk factors, early identification and precise intervention of high-risk subgroups, and risk assessment and response to public health emergencies (such as infectious diseases combined with chronic diseases).

[0074] Primary care physician / family doctor teams: Help them quickly understand the comprehensive health risk profile of contracted residents or the population in their jurisdiction, obtain personalized health management guidance and suggestions, and improve the initiative and effectiveness of services.

[0075] Research institutions and public health scholars: Provide rich multidimensional health data and analysis tools for county-level populations to study the complex relationship between socioeconomic factors and health, explore innovative public health intervention models, and assess the impact of policies.

[0076] Health management companies / smart healthcare cloud platform providers: can use this as a core technology to provide customized, refined chronic disease management solutions for counties, and expand the data-driven health governance market.

[0077] Therefore, this invention proposes a management method for population health profiles; it is understood that the computer program product is equipped with a management device for storing and executing the following method, and the management device can be implemented using a main controller, such as MCU (Micro Controller Unit), DSP (Digital Signal Processor), FPGA (Field Programmable Gate Array), SOC (System-on-Chip), etc.

[0078] As the foundational unit of my country's public health system and the core battleground for chronic disease prevention and control, the health status of residents in counties directly impacts the achievement of national health strategic goals. However, county-level chronic disease prevention and health management practices face multiple deep-seated challenges. Significant differences exist among populations in different townships and communities within counties in terms of socioeconomic background, daily behavioral patterns, environmental exposure conditions, and access to healthcare services. This high heterogeneity leads to an uneven distribution of health risk factors, and traditional management models, employing standardized interventions, cannot adapt to the individualized health needs of different groups. Health-related data is widely dispersed across multiple independent departments, including public health information systems, electronic records in medical institutions, physical examination centers, civil affairs statistics, and environmental protection. The data formats are diverse and standards inconsistent, making integration difficult. Furthermore, existing analyses rely excessively on clinical diagnosis and treatment information, failing to adequately consider key health determinants such as socioeconomic factors, behavioral characteristics, and environmental exposure levels, hindering a comprehensive understanding of the overall health of the population. The existing system primarily relies on post-hoc analysis of reported diseases or periodic health checkup data, lacking the ability to continuously track the dynamic evolution of population health status and proactively predict future health risk trends. This results in interventions often being implemented reactively only after risk events occur. Due to the lack of refined population stratification and risk assessment mechanisms, public health resource allocation is characterized by its extensive nature, making it difficult to accurately identify high-risk groups and key intervention points, leading to low resource utilization efficiency and limited prevention and control effectiveness. Adjustments to health management strategies largely depend on historical statistical reports and expert judgment, lacking the ability to dynamically optimize based on real-time data feedback, and failing to quickly respond to changes in population health status, intervention effectiveness assessments, and emerging health threats. Therefore, there is an urgent need for an innovative method that can deeply integrate multi-dimensional data, accurately characterize population health features, predict health risk evolution in real time, and achieve intelligent hierarchical management to improve the scientific rigor and adaptability of county-level chronic disease prevention and control.

[0079] Based on the above, referring to Figure 1 In one embodiment of the present invention, the method for managing population health profiles includes steps S100-S400, wherein: S100. Acquire multi-source heterogeneous data within a preset collection area, wherein the multi-source heterogeneous data includes socioeconomic dimensions, health behavior dimensions, environmental exposure dimensions, and medical service dimensions. S200. Construct feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, and generate a subgroup health profile corresponding to each feature subgroup. S300. Extract historical time-series data from the subgroup health profile, and input the historical time-series data into a preset early warning model to generate corresponding health risk trends and graded early warning signals; S400: Match the hierarchical management strategy corresponding to the health risk trend and the hierarchical early warning signal through a preset strategy knowledge base.

[0080] Multi-source heterogeneous data can be understood as a collection of data from different sources with varying structural forms. Its main function is to comprehensively reflect multi-dimensional health-related information of the population within the pre-defined collection area. Multi-source heterogeneous data can be obtained by integrating data from various channels, such as public health information systems, hospital electronic medical record systems, physical examination center records, civil affairs statistics, and environmental monitoring data. For example, socioeconomic data can be obtained from statistical yearbooks or population census data; health behavior data can be collected through questionnaires or wearable devices; environmental exposure data can be obtained from air quality monitoring stations or water quality test reports; and medical service data can be obtained from hospital information systems or community health service center records.

[0081] The process of constructing feature subgroups with different feature spectra corresponding to multi-source heterogeneous data can be understood as stratifying a population based on multi-source heterogeneous data to form groups that are internally homogeneous but differentiated from each other. This process can be achieved through clustering analysis methods, such as using K-means clustering or hierarchical clustering algorithms to process multi-source heterogeneous data and generate multiple feature subgroups. Alternatively, it can be achieved through rule-based partitioning, such as filtering and classifying multi-source heterogeneous data according to preset threshold conditions to form different feature subgroups.

[0082] The process of extracting historical time-series data from subgroup health profiles can be understood as extracting time-related continuous data from these profiles. This process can be implemented using sliding window techniques, such as setting a fixed time window to sample and extract key indicators from the subgroup health profiles. Alternatively, it can be implemented through an event-driven approach, such as triggering the extraction of relevant data when a specific event occurs.

[0083] The process of matching hierarchical management strategies with a pre-defined strategy knowledge base can be understood as retrieving the most suitable combination of management strategies from the pre-defined strategy knowledge base based on health risk trends and hierarchical early warning signals. This process can be implemented through rule-based matching, such as parsing health risk trends and hierarchical early warning signals according to pre-defined logical rules and matching corresponding strategies. Alternatively, it can be implemented through similarity calculation, such as calculating the similarity between health risk trends and hierarchical early warning signals and each entry in the strategy knowledge base, and selecting the strategy with the highest similarity as the output.

[0084] The innovation of this embodiment lies in its ability to solve the health management challenges caused by significant population heterogeneity within a county by integrating multi-source heterogeneous data to construct characteristic subgroups and generate subgroup health profiles. Simultaneously, by extracting historical time-series data and combining it with a pre-set early warning model to generate health risk trends and tiered early warning signals, it achieves a shift from passive response to proactive prediction. Furthermore, by dynamically matching tiered management strategies with a pre-set strategy knowledge base, it ensures the accuracy of resource allocation and the agility of strategy adjustments, thereby effectively addressing problems such as data silos, delayed risk identification, and inefficient resource allocation inherent in traditional management models.

[0085] The working principle of this embodiment is as follows: First, by acquiring multi-source heterogeneous data within a preset collection area, including data on socioeconomic dimensions, health behavior dimensions, environmental exposure dimensions, and medical service dimensions, a comprehensive integration of multi-dimensional data required for health management is achieved. This step overcomes the limitations of traditional analysis that relies solely on single clinical diagnosis and treatment data, and can truly reflect the overall health of the population, laying the foundation for subsequent precision management. Based on the acquired multi-source heterogeneous data, characteristic subgroups with different feature spectra are constructed, and a subgroup health profile corresponding to each characteristic subgroup is generated. The construction of characteristic subgroups addresses the significant heterogeneity of the population within a county in terms of income, education, and behavioral habits, ensuring homogeneity within subgroups while maintaining differentiation from one another, thereby providing a hierarchical basis for risk warning.

[0086] Next, historical time-series data is extracted from the generated subgroup health profiles and input into a pre-set early warning model to generate corresponding health risk trends and tiered early warning signals. By analyzing historical time-series data from subgroup health profiles, rather than overall population data, the dynamic evolution of chronic diseases such as hypertension and diabetes within specific subgroups can be accurately captured, enabling a shift from passively responding to disease reports to proactively predicting risks. Consequently, the generated health risk trends and tiered early warning signals make management more forward-looking.

[0087] Finally, a pre-set strategy knowledge base is used to match tiered management strategies corresponding to health risk trends and tiered early warning signals. This step ensures that limited medical resources are precisely targeted to high-risk subgroups, avoiding the problems of extensive resource allocation and untimely intervention in traditional models. In particular, dynamically matching strategies based on health risk trends and early warning signals, rather than relying on periodic statistical reports, makes strategy adjustments more agile and data-driven, optimizing resource allocation efficiency. In summary, this technical solution effectively addresses the core challenges of county-level health management, such as data fragmentation, diverse population needs, and delayed intervention, by systematically integrating multi-dimensional data, constructing refined characteristic subgroup profiles, dynamically predicting risk trends, and matching tiered strategies. It achieves a transformation from extensive management to precise and intelligent management.

[0088] This embodiment acquires multi-source heterogeneous data within a preset collection area, including socioeconomic, health behavior, environmental exposure, and medical service dimensions. It then constructs feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, generates subgroup health profiles for each feature subgroup, extracts historical time-series data from these profiles, and inputs this data into a preset early warning model to generate corresponding health risk trends and tiered early warning signals. Finally, a preset strategy knowledge base is used to match the tiered management strategies corresponding to the health risk trends and tiered early warning signals. In this way, by integrating multi-source heterogeneous data to construct feature subgroups and generate health profiles, and combining time-series data analysis and early warning models to achieve dynamic risk prediction, and ultimately matching tiered management strategies, this effectively solves the problems of data dispersion, high population heterogeneity, delayed risk prediction, and extensive resource allocation in county-level chronic disease prevention and control. It can deeply integrate multi-dimensional data, accurately characterize population health characteristics, predict the evolution of health risks in real time, and achieve intelligent tiered management.

[0089] Optionally, refer to Figure 2 Another embodiment of the present invention provides a method for managing population health profiles, based on the above. Figure 1 The illustrated embodiment constructs feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, and generates a subgroup health profile corresponding to each feature subgroup, including steps S210-S240, wherein: S210. Construct a comprehensive quantitative indicator system covering each data point in the multi-source heterogeneous data; S220. The comprehensive quantitative index system is subjected to feature dimensionality reduction processing by principal component analysis or factor analysis to extract multiple potential core common factors that can characterize the determinants of population health. S230. Based on clustering algorithms and multiple potential core common factors, the population within the preset collection area is divided into layers to generate multiple feature subgroups that are internally homogeneous but different from each other. S240. Generate a multi-dimensional subgroup health profile corresponding to each of the aforementioned feature subgroups. The multi-dimensional subgroup health profile includes the distribution of key feature indicators of the corresponding subgroup, a list of major health problems, analysis of potential risk drivers, and preliminary intervention direction suggestions.

[0090] The comprehensive quantitative indicator system refers to the unified transformation of multi-source heterogeneous data, including socioeconomic, health behavior, environmental exposure, and medical service dimensions, into a quantifiable standard system. This can be achieved using standardized scoring methods, weighted index methods, or analytic hierarchy process (AHP). The aim is to ensure that all health determinants are included in the analytical framework, providing a comprehensive and consistent data foundation for subsequent processing. Potential core common factors can be understood as core variables that reflect the essential characteristics of population health determinants, extracted after dimensionality reduction of the comprehensive quantitative indicator system. These can be achieved through principal component analysis, factor analysis, or independent component analysis, aiming to eliminate redundancy and noise in the original data and focus on the core variables affecting health. In practical applications, clustering algorithms refer to methods for stratifying populations based on the dimensionality-reduced potential core common factors. These can be implemented using algorithms such as K-means clustering, hierarchical clustering, or density clustering. The goal is to generate multiple internally homogeneous but mutually differentiated characteristic subgroups, thereby accurately capturing the heterogeneity of county-level populations in terms of socioeconomic status, behavioral habits, and environmental exposure.

[0091] The aforementioned technical solution effectively addresses the problems of insufficient integration of multidimensional health data and inaccurate subgroup segmentation by systematically integrating multi-source heterogeneous data and implementing feature dimensionality reduction and cluster analysis. First, a comprehensive quantitative indicator system covering every data point in the multi-source heterogeneous data is constructed, unifying scattered dimensions such as socioeconomic status, health behavior, environmental exposure, and medical services into quantifiable standards. This avoids the limitations of traditional analysis relying solely on clinical data, providing a comprehensive and consistent data foundation for subsequent processing. Based on this, principal component analysis or factor analysis is used to perform feature dimensionality reduction on the comprehensive quantitative indicator system, extracting multiple potential core common factors that characterize population health determinants. This effectively eliminates redundancy and noise in the original data, enabling the potential core common factors to more accurately reflect the essential characteristics of population health determinants. Specifically, based on clustering algorithms and multiple potential core common factors, the population within a pre-defined collection area is stratified. The dimensionality-reduced core common factors are used as input to generate multiple internally homogeneous but mutually differentiated feature subgroups. This ensures that the segmentation process closely revolves around key health-related factors, thereby achieving a high degree of consistency in health characteristics among members within subgroups and significant differences between subgroups. Finally, a multi-dimensional subgroup health profile is generated for each characteristic subgroup. Combining the distribution of key characteristic indicators, a list of major health problems, analysis of potential risk drivers, and preliminary intervention direction suggestions, it not only comprehensively describes the current health status of the subgroup, but also deeply analyzes the causes of risks and provides targeted intervention clues. This provides a direct basis for subsequent risk warning and strategy formulation, and promotes the transformation of health management from extensive to intensive.

[0092] The above technical solutions enable the systematic integration and feature extraction of multi-source heterogeneous data, solving the problem that subgroup classification cannot accurately reflect the heterogeneity of population health determinants, and laying a solid foundation for precision health management.

[0093] Optionally, refer to Figure 3 Another embodiment of the present invention provides a method for managing population health profiles, based on the above. Figure 2 The illustrated embodiment uses a clustering algorithm and multiple potential core common factors to stratify the population within the preset collection area, generating multiple feature subgroups that are internally homogeneous but differentiated from each other, including steps S231-S234, wherein: S231. A prototype clustering algorithm capable of simultaneously processing numerical continuous variables and categorical discrete variables is adopted as the basic partitioning algorithm. S232. Based on the data characteristics of the potential core common factors, construct a hybrid dissimilarity measurement function suitable for the prototype clustering algorithm; S233. The partitioning effect of the prototype clustering algorithm under different preset cluster numbers is iteratively verified by using a combination of silhouette coefficient and Davidson-Bolding index as evaluation indicators to determine the optimal number of feature subgroups. S234. Use a density-based noise spatial clustering algorithm to optimize the boundary of the partitioning results of the prototype clustering algorithm, identify and exclude discrete noise points, and generate the feature subgroup.

[0094] Among them, the prototype clustering algorithm refers to a clustering method that can simultaneously handle numerical continuous variables and categorical discrete variables. It can be implemented using the K-prototype algorithm or an extended fuzzy C-means algorithm, aiming to overcome the incompatibility of traditional clustering algorithms with mixed data types, thereby ensuring the uniformity and accuracy of the clustering process. The mixed dissimilarity metric function is a comprehensive function combining distance calculation for numerical variables and similarity assessment for categorical variables. It can be implemented by combining weighted Euclidean distance and Jaccard similarity coefficient, aiming to reflect the intrinsic correlation between multidimensional health determinants and improve the representativeness of clustering results. The combined evaluation index of silhouette coefficient and Davidson-Bolding index is a comprehensive evaluation tool for measuring clustering effectiveness. It can be implemented by calculating the ratio of intra-cluster compactness to inter-cluster separation, aiming to provide objective quantitative basis for determining the optimal number of clusters and avoiding bias caused by subjective experience. Density-based noise spatial clustering is a method that uses the density distribution characteristics of data points to optimize boundaries. It can be implemented using the DBSCAN or OPTICS algorithms. Its purpose is to identify and eliminate discrete noise points, ensuring that the generated feature subgroups have clear boundaries and significant differentiation.

[0095] The aforementioned technical solution addresses the core challenge of accurately segmenting characteristic subgroups under multi-source heterogeneous health data by systematically improving the clustering process. First, the prototype clustering algorithm, considering the coexistence of socioeconomic and health behavior dimensions in county-level population data, ensures unified processing of numerical and categorical variables, avoiding subgroup structure distortion caused by data type fragmentation. Second, the mixture dissimilarity metric function, through customized distance calculation rules, is compatible with both linear changes in continuous variables and categorical differences in discrete variables, significantly improving the accuracy of clustering results in representing population health characteristics. Furthermore, the combined evaluation index of silhouette coefficient and Davidson-Bolding index, through iterative verification of the segmentation effect under different preset clustering numbers, achieves objective quantification of clustering number selection, overcoming the limitations of a single index. Finally, the density-based noise-applied spatial clustering algorithm effectively eliminates the interference of data noise on subgroup boundaries by optimizing the partitioning results, ensuring that the generated characteristic subgroups maintain a high degree of consistency in internal health characteristics while reflecting significant differences between different subgroups.

[0096] Furthermore, the aforementioned technical solution is closely integrated with steps such as acquiring multi-source heterogeneous data, constructing a comprehensive quantitative indicator system, and extracting potential core common factors, forming a complete technical chain. Through this chain, not only can the health heterogeneity of the county population be accurately characterized, but reliable data support is also provided for subsequent health risk trend prediction and the matching of hierarchical management strategies, thereby significantly improving the scientific nature and effectiveness of chronic disease prevention and control.

[0097] Optionally, refer to Figure 4 Another embodiment of the present invention provides a method for managing population health profiles, based on the above. Figure 1 The embodiment shown extracts historical time-series data from the subgroup health profile and inputs the historical time-series data into a preset early warning model to generate corresponding health risk trends and graded early warning signals, including steps S310-S350, wherein: S310. Extract the first observation of key health outcome indicators and the second observation of core risk factor exposure levels at consecutive time points from the subgroup health profile of each subgroup. S320. Construct a multivariate time series dataset corresponding to the first observation and the second observation; S330. Perform sliding window processing, seasonal difference and stationarity test on each data point of the multivariate time series dataset to generate the corresponding historical time series data. S340. Input the historical time series data into the preset early warning model and output the health risk trend within a specific future time span; S350. Compare the health risk trend with the preset multi-level risk thresholds to generate the graded early warning signal.

[0098] Key health outcome indicators refer to core outcome indicators that directly reflect the health status of a population. These can be prevalence of chronic diseases, mortality rates, or incidence of specific health events, aiming to focus on core variables affecting health. Core risk factor exposure levels refer to key behavioral, environmental, or socioeconomic factors that may significantly affect health outcomes. These can be smoking rates, obesity rates, or pollution exposure levels, aiming to capture the main drivers of health changes. Multivariate time series datasets are collections of data integrating multiple time-varying variables. They can be implemented in matrix or tensor form, aiming to comprehensively reflect the dynamic relationship between health outcomes and risk factors. Sliding window processing is a technique for capturing short-term fluctuation patterns. It can be implemented using fixed or dynamic windows, aiming to improve sensitivity to local changes. Seasonal differencing is a method for eliminating periodic effects. It can be implemented through differencing operations or Fourier transforms, aiming to remove periodic interference from the data. Stationarity testing is a technique for verifying whether time series data meets modeling requirements. It can be implemented using ADF tests or KPSS tests, aiming to ensure that the data is suitable for predictive modeling.

[0099] The aforementioned scheme achieves precise dynamic risk prediction through a series of steps. First, continuous observations of key health outcome indicators and core risk factor exposure levels are extracted from subgroup health profiles. This process focuses on highly correlated dynamic indicators, avoiding interference from irrelevant data and laying the foundation for subsequent analysis. Next, a multivariate time-series dataset is constructed, integrating the temporal correlation between health outcomes and risk factors, considering their dynamic interaction, and thus comprehensively reflecting the evolution mechanism of health risk. Subsequently, by performing sliding window processing, seasonal differencing, and stationarity tests on the data, the issues of data non-stationarity and seasonality are resolved, generating high-quality historical time-series data that provides stable and reliable input for the early warning model. Based on this, the processed data is input into a pre-set early warning model, using the rigorously processed data to drive the model's output of future health risk trends, effectively avoiding prediction bias caused by noise in the original data. Finally, by comparing health risk trends with pre-set multi-level risk thresholds, a tiered early warning signal is generated, ensuring that the early warning signal is based on clear risk grading standards, is operable, and can guide precise intervention measures at different risk levels. The above solution, combined with the aforementioned acquisition of multi-source heterogeneous data, construction of characteristic subgroups, and generation of subgroup health profiles, forms a complete chain from data collection to risk warning, significantly improving the scientific nature and effectiveness of health management.

[0100] Optionally, refer to Figure 5The present invention also provides a method for managing population health profiles, based on the above. Figure 4 The illustrated embodiment extracts the first observations of key health outcome indicators and the second observations of core risk factor exposure levels at consecutive time points from the subgroup health profile of each subgroup, including steps S311-S312, wherein: S311. From the list of major health problems in the subgroup health profile, extract the prevalence of hypertension, the incidence of diabetes and the mortality rate of cardiovascular and cerebrovascular diseases as the first observations of key health outcome indicators. S312. From the distribution of key feature indicators of the subgroup health profile, extract the smoking rate of the adult population, the average body mass index, and the participation rate in moderate to high intensity physical exercise as the second observation of the exposure level of core risk factors.

[0101] Key health outcome indicators refer to core quantitative results that directly reflect the health status of the population. These can be achieved using indicators with clear clinical significance, such as the prevalence of hypertension, the incidence of diabetes, and the mortality rate from cardiovascular and cerebrovascular diseases. The aim is to ensure that the selected indicators are highly representative and closely related to the county's chronic disease prevention and control goals. Core risk factor exposure levels refer to the behavioral and environmental exposure-related drivers that affect health outcomes. These can be achieved using indicators that have been validated as key variables, such as the adult population smoking rate, average body mass index, and participation rate in moderate-to-vigorous intensity physical activity. The aim is to comprehensively identify the potential sources of health risks in subgroups.

[0102] The aforementioned method extracts the prevalence of hypertension, the incidence of diabetes, and the mortality rate of cardiovascular and cerebrovascular diseases from a list of major health problems as the first observations. This ensures that the selection of these indicators is based on a targeted summary of health problems generated from subgroup feature spectra, thus avoiding data redundancy and loss of clinical significance caused by arbitrary indicator selection. Simultaneously, it extracts the adult population smoking rate, average body mass index, and participation rate in moderate-to-vigorous intensity physical exercise as the second observations from the distribution of key feature indicators. Utilizing the feature dimensionality reduction results of multi-source heterogeneous data, it systematically captures the impact of behavioral habits and environmental exposures on health. Through this structured indicator definition approach, the generation process of historical time-series data is standardized, allowing the input of the early warning model to focus on highly correlated data, thereby supporting the accuracy of health risk trend prediction and the targeting of tiered early warning signals. Furthermore, combining this method with the aforementioned technical solutions for constructing feature subgroups and generating subgroup health profiles further improves the consistency and representativeness of data collection, solves the input inconsistency problem caused by ambiguity in indicator selection, and significantly enhances the scientific rigor and reliability of health risk prediction.

[0103] Optionally, refer to Figure 6Another embodiment of the present invention provides a method for managing population health profiles, based on the above. Figure 5 The illustrated embodiment of the population health profile management method further includes steps S500-S800, wherein: S500. Construct a neural network infrastructure with a long short-term memory network or a gated recurrent unit as its core. The neural network infrastructure is used to capture long-term dependencies in time series. S600. An attention mechanism is built into the neural network infrastructure. The attention mechanism is used to dynamically evaluate the importance of the hidden state at different historical time points to the current prediction task during the forward computation process and generate the corresponding weight allocation result. S700. The weighted fusion of all hidden states output by the neural network infrastructure is performed using the weight allocation result. A multi-objective loss function composed of mean squared error loss function and mean absolute percentage error loss function is defined. An adaptive moment estimation optimization algorithm is used to train the entire network parameters, including the attention mechanism, end-to-end. S800. During model training, an independent validation dataset is used to evaluate the model's prediction accuracy, and models with a root mean square error of less than a preset threshold in the test set are saved as the preset early warning models.

[0104] Long Short-Term Memory (LSTM) networks, or gated recurrent units, are deep learning model structures specifically designed for processing time-series data. They effectively model information spanning long time periods through internal memory units and gating mechanisms. In practice, standard LSM networks, bidirectional LSM networks, or stacked gated recurrent units can be used to overcome the difficulty of traditional models in capturing long-range dependencies in time series data. Attention mechanisms can be understood as dynamic weight allocation modules that calculate the correlation score between the hidden state at each historical time point and the current prediction task, generating a corresponding temporal weight distribution. In practice, dot-product attention, additive attention, or multi-head attention can be used to improve the model's ability to focus on key time segments. Furthermore, multi-objective loss functions are composite optimization objectives that consider multiple error metrics simultaneously. They can be implemented through linear weighting, dynamic weight adjustment, etc., aiming to improve the model's robustness to different types of prediction errors.

[0105] The overall operational logic of this scheme is as follows: First, a neural network infrastructure built upon Long Short-Term Memory (LSTM) networks or gated recurrent units (GRUs) enables in-depth analysis of the correlations between distant data points in time series, laying a structural foundation for subsequent risk trend prediction. Building upon this, a built-in attention mechanism allows the model to dynamically evaluate the importance of hidden states at different historical time points during forward computation, generating targeted weight allocations to avoid missing crucial information due to fixed weighting. Subsequently, the generated weight allocations are used to weight and fuse all hidden states output by the neural network infrastructure, ensuring that critical historical information is prioritized and utilized. Simultaneously, a multi-objective loss function consisting of a mean squared error loss function and a mean absolute percentage error loss function is defined, and end-to-end training is performed using an adaptive moment estimation (IME) optimization algorithm. This ensures that the model balances the absolute bias and relative proportion of prediction errors during optimization. Finally, during model training, an independent validation dataset is used to evaluate prediction accuracy, and only models with a root mean square error (RMSE) below a preset threshold are saved. This rigorous validation mechanism selects models with strong generalization capabilities. The organic combination of these technologies effectively solves the problems that models struggle to capture long-term dependencies in time-series data and cannot dynamically identify the critical impact of different time points in historical data on the current prediction task, significantly improving the accuracy and adaptability of health risk prediction.

[0106] Optionally, refer to Figure 7 Another embodiment of the present invention provides a method for managing population health profiles, based on the above. Figure 1 The illustrated embodiment matches a tiered management strategy corresponding to the health risk trend and the tiered early warning signal using a preset strategy knowledge base, including steps S410-S430, wherein: S410. Generate a strategy query request corresponding to the health risk trend and the graded early warning signal; S420. Input the strategy query request into the preset strategy knowledge base for matching and retrieval, and determine one or more most suitable hierarchical management strategy combinations; S430. Perform resource consumption simulation and preliminary feasibility assessment on the combination of hierarchical management strategies, and output the hierarchical management strategies.

[0107] The strategy query request refers to a structured query instruction generated based on current health risk trends and tiered early warning signals. It can be implemented using data formats such as JSON and XML, aiming to ensure that the query content accurately maps to the dynamic characteristics of health risks and avoids generalized matching that could lead to a disconnect between strategies and risk levels. The preset strategy knowledge base can be understood as a database system storing the rules and logic of tiered management strategies. It can be implemented using a relational database or a graph database, aiming to support automated strategy selection based on risk characteristics. Resource consumption simulation refers to the process of quantitatively estimating the resources required for the implementation of tiered management strategies. It can be implemented using linear programming models or Monte Carlo simulation methods, aiming to identify potential resource bottlenecks and adjust strategy combinations to ensure executability.

[0108] This solution generates strategy query requests, enabling the query content to accurately reflect the dynamic characteristics of health risk trends and tiered early warning signals, thus avoiding strategy bias caused by generalized matching. Subsequently, the strategy query requests are input into a pre-set strategy knowledge base for matching and retrieval. This process, based on the structured representation of risk characteristics, automates strategy screening, ensuring the targeting and adaptability of the screening results. Building on this, resource consumption simulation and preliminary feasibility assessment are conducted on the combination of tiered management strategies. Combined with real-time distribution and capacity data of medical and health resources, the resource requirements for strategy implementation are quantitatively estimated and dynamically compared with the actual resource status. Potential bottlenecks are identified, and the strategy combination is adjusted. This process not only solves the problem of mismatch between strategies and resource capabilities but also extends from theoretical matching to resource adaptation, ensuring the feasibility and accuracy of management measures under the limited resource conditions of a county.

[0109] The above technical solutions effectively solve the problem of resource allocation imbalance that may result from relying solely on static matching results from the knowledge base, and improve the adaptability and execution efficiency of the hierarchical management strategy in the prevention and control of chronic diseases in counties.

[0110] Optionally, refer to Figure 8 Another embodiment of the present invention provides a method for managing population health profiles, based on the above. Figure 7 The illustrated embodiment performs resource consumption simulation and preliminary feasibility assessment on the combination of hierarchical management strategies, and outputs the hierarchical management strategy, including steps S431-S434, wherein: S431. Obtain real-time distribution data and available capacity data of medical and health resources within the preset collection area; S432. Quantitatively estimate the various resources required for each measure in the hierarchical management strategy combination by using a preset resource consumption simulation model, and generate corresponding resource demand estimation results. S433. Compare and analyze the resource demand estimation results with real-time distributed data and available capacity data respectively to identify potential resource bottlenecks and conflicts, and determine the comparison and analysis results. S434. Based on the comparison and analysis results, adjust and optimize the combination of hierarchical management strategies to generate the hierarchical management strategy.

[0111] Real-time distribution data refers to the actual distribution of medical and health resources within a pre-defined collection area at a specific moment. This can be achieved through IoT sensors, real-time monitoring systems, or online reporting platforms. Available capacity data can be understood as the maximum carrying capacity of medical and health resources within the pre-defined collection area. This data can be obtained through historical data analysis, equipment specifications, or personnel scheduling, with the aim of providing accurate baseline information for resource assessment.

[0112] In practical applications, a pre-set resource consumption simulation model is a tool that can transform intervention measures in a combination of hierarchical management strategies into resource demand calculations. It can be implemented using methods such as linear programming models, Monte Carlo simulations, or discrete event simulations. The comparative analysis results can be quantitative indicators of resource supply and demand, such as resource utilization rate, remaining capacity, or gap ratio, aiming to systematically reveal potential problems in resource allocation.

[0113] This scheme first ensures the dynamism and accuracy of the basic data for resource assessment by acquiring real-time distribution and available capacity data within a pre-defined collection area, thus avoiding biases caused by relying on static data. Next, a pre-defined resource consumption simulation model is used to quantitatively estimate the various measures in the tiered management strategy combination, transforming abstract intervention measures into specific resource demand indicators. This process replaces traditional subjective experience-based judgments, improving the objectivity and repeatability of the estimation. Subsequently, by comparing and analyzing the resource demand estimation results with real-time distribution and available capacity data, resource bottlenecks and conflicts can be accurately identified, such as potential shortages of medical equipment or overburdened human resources in certain areas. Finally, based on the comparative analysis results, the tiered management strategy combination is adjusted and optimized to ensure that the final output tiered management strategy both meets resource constraints and efficiently responds to health risks.

[0114] The aforementioned solution achieves dynamic integration from data acquisition to strategy optimization through a closed-loop design of the entire resource assessment process. For example, in the context of chronic disease prevention and control, when a township health center is found to have insufficient drug inventory, the priority of intervention measures can be adjusted or resources can be reallocated to other areas, thereby avoiding resource shortages or waste. Furthermore, combining this solution with the aforementioned population health profiling management method can further improve the accuracy and efficiency of resource allocation, making the tiered management strategy more adaptable and implementable.

[0115] The present invention also proposes a management device, the management device comprising: a memory, a processor, and a management program for a population health profile stored in the memory and executable on the processor, the management program for the population health profile being configured to implement the population health profile management method as described above.

[0116] The core innovation of this embodiment lies in combining multi-source heterogeneous data with a dynamic risk early warning mechanism through a closed-loop collaborative design, and introducing a precise hierarchical strategy matching method based on feature subgroups. This effectively solves problems such as severe data silos, lagging risk prediction, and extensive resource allocation in the county-level public health system. This solution not only achieves a shift from passive response to proactive prediction, but also significantly improves the scientific nature and resource utilization efficiency of chronic disease prevention and control.

[0117] The memory is used to centrally store multi-source heterogeneous data within a preset collection area, including socioeconomic dimensions, health behavior dimensions, environmental exposure dimensions, and medical service dimensions, as well as feature subgroup information, historical time-series data, and strategy knowledge base content. This ensures unified access and rapid retrieval of cross-departmental data, avoiding data silos between independent systems such as public health information systems and hospital HIS / EMR systems, and providing a complete data foundation for subsequent analysis. The processor performs core tasks such as data processing, feature dimensionality reduction, clustering calculation, and early warning model inference. Through its high-efficiency computing power, it processes complex operations such as principal component analysis, prototype clustering algorithms, and long short-term memory networks in real time, transforming health outcome indicators and risk factor exposure levels at continuous time points into dynamic risk trends. This overcomes the lag inherent in traditional methods that rely on annual physical examination data, enabling proactive prediction of key indicators such as hypertension prevalence and diabetes incidence. The population health profile management program generates characteristic subgroup health profiles based on multi-source heterogeneous data in the memory, and outputs tiered early warning signals based on historical time-series data input into the early warning model. Then, it outputs tiered management strategies by matching resource consumption simulation results with a strategy knowledge base. This design enables the program to dynamically adjust the intervention direction based on the distribution of key characteristic indicators and potential risk drivers of subgroups, ensuring that resource allocation accurately matches the needs of high-risk subgroups. At the same time, through comparative analysis of real-time medical and health resource distribution data, the strategy adjustment has data-driven agility and adapts to rapid changes in the population's health status.

[0118] This embodiment proposes a method for managing population health profiles, including acquiring multi-source heterogeneous data within a preset collection area, constructing feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, and generating a subgroup health profile corresponding to each feature subgroup; extracting historical time-series data from the subgroup health profiles and inputting the historical time-series data into a preset early warning model to generate corresponding health risk trends and graded early warning signals; and matching graded management strategies corresponding to the health risk trends and graded early warning signals through a preset strategy knowledge base. Multi-source heterogeneous data can be understood as a collection of data from different sources with varying structural forms, its main function being to comprehensively reflect multi-dimensional information related to population health within the preset collection area. For example, socioeconomic data can be obtained through statistical yearbooks or population census data; health behavior data can be collected through questionnaires or wearable devices; environmental exposure data can be obtained through air quality monitoring stations or water quality testing reports; and medical service data can be obtained through hospital information systems or community health service center records.

[0119] Furthermore, the process of constructing feature subgroups can be achieved through clustering analysis methods, such as using K-means clustering or hierarchical clustering algorithms to process multi-source heterogeneous data, thereby generating multiple feature subgroups. The process of extracting historical time-series data from the subgroup health profiles can be achieved through sliding window techniques, such as setting a fixed time window to sample and extract key indicators from the subgroup health profiles. The process of matching hierarchical management strategies with a pre-defined strategy knowledge base can be achieved through rule matching or similarity calculation. For example, health risk trends and hierarchical warning signals can be parsed and matched with corresponding strategies according to pre-defined logical rules, or the similarity between health risk trends and hierarchical warning signals and each entry in the strategy knowledge base can be calculated, and the strategy with the highest similarity can be selected as the output result.

[0120] This embodiment addresses the health management challenges caused by significant population heterogeneity within a county by integrating multi-source heterogeneous data to construct characteristic subgroups and generate subgroup health profiles. Simultaneously, by extracting historical time-series data and combining it with a pre-set early warning model to generate health risk trends and tiered early warning signals, it achieves a shift from passive response to proactive prediction. Furthermore, by dynamically matching tiered management strategies with a pre-set strategy knowledge base, it ensures the accuracy of resource allocation and the agility of strategy adjustments, effectively addressing issues such as data silos, delayed risk identification, and inefficient resource allocation inherent in traditional management models.

[0121] It is worth noting that since the management device of the present invention is based on the above-mentioned management method of population health profile, the embodiments of the management device of the present invention include all the technical solutions of all embodiments of the above-mentioned management method of population health profile, and the technical effects achieved are exactly the same, so they will not be repeated here.

[0122] The present invention also proposes a computer program product, which includes the management device as described in the above embodiments.

[0123] The core innovation of this embodiment lies in the integrated combination of memory, processor, and management program to construct a complete execution environment, thereby solving the problems of data silos and dimensional limitations. Specifically, the memory persistently stores multi-source heterogeneous data and model parameters, enabling efficient retrieval of historical time-series data to support risk trend prediction; the processor performs feature dimensionality reduction, clustering, and early warning model calculations, ensuring real-time extraction of potential core common factors and generation of subgroup health profiles, overcoming the challenges of delayed risk identification and passive response; the management program is configured to manage the health profiles of the population, automating the matching process between tiered early warning signals and the strategy knowledge base, avoiding the shortcomings of inefficient and inefficient resource allocation. Furthermore, the computer program product, as a software distribution carrier, facilitates rapid installation and iteration in county-level computing environments, ensuring that the management method can adapt to dynamically changing health data and intervention needs, solving the problem of lack of data-driven and agile strategy adjustments.

[0124] Through the above technical solution, this embodiment realizes the transformation from passive response to proactive prediction, while ensuring the accuracy of resource allocation and the agility of strategy adjustment. It effectively addresses the core challenges of traditional management models, such as data fragmentation, diverse population needs, and delayed intervention, and promotes the transformation of county-level health management from extensive management to precise and intelligent management.

[0125] It is worth noting that, since the computer program product of the present invention is based on the above-mentioned management device, the embodiments of the computer program product of the present invention include all the technical solutions of all the embodiments of the above-mentioned management device, and the technical effects achieved are exactly the same, which will not be repeated here.

[0126] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0127] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0128] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0129] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for managing population health profiles, characterized in that, The management methods for the aforementioned population health profiles include: Acquire multi-source heterogeneous data within a preset collection area, the multi-source heterogeneous data including socioeconomic dimensions, health behavior dimensions, environmental exposure dimensions, and medical service dimensions; Construct feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, and generate a subgroup health profile corresponding to each feature subgroup; Historical time-series data are extracted from the subgroup health profiles and input into a preset early warning model to generate corresponding health risk trends and graded early warning signals. A hierarchical management strategy is matched with the health risk trend and the hierarchical early warning signal by a preset strategy knowledge base.

2. The method for managing population health profiles as described in claim 1, characterized in that, The step of constructing feature subgroups with different feature spectra corresponding to the multi-source heterogeneous data, and generating a subgroup health profile corresponding to each feature subgroup, includes: Construct a comprehensive quantitative indicator system covering every data point in the multi-source heterogeneous data; The comprehensive quantitative index system is subjected to feature dimensionality reduction processing by principal component analysis or factor analysis to extract multiple potential core common factors that can characterize the determinants of population health. Based on clustering algorithms and multiple potential core common factors, the population within the preset collection area is stratified and divided into multiple feature subgroups that are internally homogeneous but different from each other. Generate a multi-dimensional subgroup health profile corresponding to each of the aforementioned feature subgroups. The multi-dimensional subgroup health profile includes the distribution of key feature indicators of the corresponding subgroup, a list of major health problems, analysis of potential risk drivers, and preliminary intervention direction suggestions.

3. The method for managing population health profiles as described in claim 2, characterized in that, The method involves stratifying the population within the preset collection area based on a clustering algorithm and multiple potential core common factors, generating multiple feature subgroups that are internally homogeneous but differentiated from each other, including: A prototype clustering algorithm capable of simultaneously processing numerical continuous variables and categorical discrete variables is adopted as the basic partitioning algorithm; Based on the data characteristics of the potential core common factors, a hybrid dissimilarity measurement function suitable for the prototype clustering algorithm is constructed; The prototype clustering algorithm was iteratively verified to determine the optimal number of feature subgroups by using a combination of silhouette coefficient and Davidson-Bolding index as evaluation metrics. The partitioning results of the prototype clustering algorithm are optimized by using a density-based noise spatial clustering algorithm to identify and exclude discrete noise points, thereby generating the feature subgroup.

4. The method for managing population health profiles as described in claim 1, characterized in that, The step of extracting historical time-series data from the subgroup health profile and inputting the historical time-series data into a preset early warning model to generate corresponding health risk trends and graded early warning signals includes: Extract the first observation of key health outcome indicators and the second observation of core risk factor exposure levels at consecutive time points from the subgroup health profile of each subgroup; Construct a multivariate time series dataset corresponding to the first observation and the second observation; For each data point in the multivariate time series dataset, sliding window processing, seasonal differencing, and stationarity tests are performed to generate corresponding historical time series data. The historical time series data is input into the preset early warning model, and the health risk trend within a specific future time span is output. The health risk trend is compared with the preset multi-level risk thresholds to generate the graded early warning signal.

5. The method for managing population health profiles as described in claim 4, characterized in that, The extraction of the first observation of key health outcome indicators and the second observation of core risk factor exposure levels at consecutive time points from the subgroup health profile of each subgroup includes: From the list of major health problems in the subgroup health profile, the prevalence of hypertension, the incidence of diabetes, and the mortality rate of cardiovascular and cerebrovascular diseases were extracted as the first observations of key health outcome indicators. From the distribution of key characteristic indicators of the subgroup health profile, the smoking rate of the adult population, the average body mass index, and the participation rate in moderate to high intensity physical exercise were extracted as the second observation of the exposure level of core risk factors.

6. The method for managing population health profiles as described in claim 5, characterized in that, The management method for the population health profile also includes: Construct a neural network infrastructure with a long short-term memory network or gated recurrent unit as its core, the neural network infrastructure being used to capture long-term dependencies in time series; An attention mechanism is incorporated into the neural network infrastructure. This attention mechanism is used to dynamically evaluate the importance of the hidden state at different historical time points to the current prediction task during the forward computation process and generate corresponding weight allocation results. The weighted fusion of all hidden states output by the neural network infrastructure is performed using the weighted allocation results. A multi-objective loss function consisting of mean squared error loss function and mean absolute percentage error loss function is defined. An adaptive moment estimation optimization algorithm is used to train the entire network parameters, including the attention mechanism, end-to-end. During model training, an independent validation dataset is used to evaluate the model's prediction accuracy, and models with a root mean square error in the test set below a preset threshold are saved as the preset early warning models.

7. The method for managing population health profiles as described in claim 1, characterized in that, The step of matching the hierarchical management strategy corresponding to the health risk trend and the hierarchical early warning signal with a preset strategy knowledge base includes: Generate a strategy query request corresponding to the health risk trend and the graded early warning signal; The strategy query request is input into the preset strategy knowledge base for matching and retrieval to determine one or more most suitable combinations of hierarchical management strategies; Resource consumption simulation and preliminary feasibility assessment are performed on the combination of hierarchical management strategies, and the hierarchical management strategies are output.

8. The method for managing population health profiles as described in claim 7, characterized in that, The process of simulating resource consumption and conducting a preliminary feasibility assessment of the combination of hierarchical management strategies, and outputting the hierarchical management strategies, includes: Acquire real-time distribution data and available capacity data of medical and health resources within the preset collection area; The resource consumption simulation model is used to quantitatively estimate the various resources required for each measure in the hierarchical management strategy combination, and the corresponding resource demand estimation results are generated. The resource demand estimation results are compared and analyzed with real-time distributed data and available capacity data to identify potential resource bottlenecks and conflicts, and to determine the comparison and analysis results. Based on the comparison and analysis results, the hierarchical management strategy combination is adjusted and optimized to generate the hierarchical management strategy.

9. A management device, characterized in that, The management device includes: a memory, a processor, and a management program for a population health profile stored in the memory and executable on the processor, wherein the management program for the population health profile is configured to implement the management method for the population health profile as described in any one of claims 1 to 8.

10. A computer program product, characterized in that, Includes the management device as described in claim 9.

Images