System for analyzing interrelationship among treatable mortality, healthcare provision, and urban shrinkage using a panel vector autoregression approach

The PVAR model addresses the limitations of existing systems by dynamically analyzing treatable mortality and medical supply interactions, enhancing policy reliability and prediction accuracy through OIRF and FEVD visualization.

KR103044692B1Active Publication Date: 2026-11-11SEOUL NATIONAL UNIVERSITY R&DB FOUNDATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
KR1020250163322
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Priority Date
2025-08-29
Filing Date
2025-11-03
Publication Date
2026-11-11
Estimated Expiration
2045-11-03

AI Technical Summary

Technical Problem

Existing analysis systems fail to provide a comprehensive, dynamic framework for understanding the bidirectional interactions between treatable mortality, medical supply, and regional extinction, neglecting regional heterogeneity and time-series changes, and lack the ability to quantify policy interventions effectively.

Method used

A Panel Vector Autoregression (PVAR) model is employed to analyze the dynamic interactions between treatable mortality, medical supply, and regional extinction, using Orthogonal Impulse Response Functions (OIRF) and Forecast Error Variance Decomposition (FEVD) to visualize and quantify the influence between variables.

Benefits of technology

The system provides a statistical framework to interpret complex relationships between medical resources and population changes, enabling early diagnosis of medical supply shortages and population outflow, improving policy reliability and prediction accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure R1020250163322_ABST
    Figure R1020250163322_ABST
Patent Text Reader

Abstract

A system for analyzing dynamic interactions between treatable mortality, medical supply, and regional extinction using a panel vector autoregressive model approach is disclosed. The system for analyzing dynamic interactions between treatable mortality, medical supply, and regional extinction according to an embodiment of the present invention comprises: a variable value collection unit that extracts pre-set variable values ​​for a pre-set investigation period and a target group and structures them in the form of a balanced panel; a bidirectional relationship review unit that evaluates the dynamic interdependence between multiple time series of the pre-set variable values ​​using a statistical method utilizing a PVAR model (panel vector autoregressive model) to derive an estimation result including a bidirectional relationship; and a derived data output unit that processes the estimation result derived through the bidirectional relationship review unit using Orthogonalized Impulse Response Functions (OIRF) and Forecast Error Variance Decomposition (FEVD) and then outputs the result.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to a dynamic interaction analysis system, and more specifically, to a dynamic interaction analysis system between treatable mortality, medical supply, and fat depletion using a panel vector autoregression model approach. Background Technology

[0002] The provision of medical services in cities and communities has long been regarded as a key indicator of public welfare, and it has a structure in which accessibility and quality vary significantly depending on population distribution, economic conditions, and administrative support systems. In particular, the rapid urban concentration that has occurred over the past few decades has simultaneously led to an oversupply of medical infrastructure in metropolitan areas and a reduction in medical infrastructure in small and medium-sized cities in the provinces. This imbalance has structurally deepened disparities in survival rates and health between regions, and as a result, the spatial allocation of medical resources has emerged as a key determinant of population health.

[0003] In this context, 'treatable mortality' has been utilized as a representative indicator for evaluating the quality of public health. However, existing studies have primarily focused on the simple correlation between the level of medical services and mortality rates at a single point in time or in a single region, making it difficult to reflect long-term changes or interactions between regions. For instance, phenomena such as the delayed impact of increased medical facilities on actual mortality reductions, or conversely, the acceleration of population outflow in regions lacking medical resources, were difficult to explain using existing simple regression models.

[0004] Furthermore, despite the fact that the phenomenon of regional extinction is not merely a matter of population decline but a complex crisis leading to the simultaneous collapse of social capital and infrastructure, attempts to analyze it in conjunction with the supply structure of medical services and health indicators have been very rare. Most policy studies have limited their treatment of regional extinction to administrative and financial issues, failing to empirically verify the interaction between population influx and medical supply, as well as the resulting changes in public health levels.

[0005] Meanwhile, regional imbalances are intensifying regarding the number of doctors and hospital beds, which are key elements of medical supply. The concentration of medical personnel in large hospitals in the capital area has exacerbated staffing shortages at regional medical institutions, leading to an increase in cases in small and medium-sized cities where even treatable diseases cannot be addressed in a timely manner. However, it is also pointed out that simply increasing the number of personnel is insufficient to solve the problem. Existing models have failed to encompass the simultaneous interaction of complex factors, such as the qualitative distribution of medical personnel, the patient-to-bed ratio, and access to emergency medical services.

[0006] Traditional health economics approaches have primarily utilized cross-sectional regression models or fixed-effects panel models to identify causal relationships between medical resource inputs and health outcomes. However, these approaches have limitations in explaining complex systems where mutual influences among variables occur simultaneously. In particular, since the relationships between medical supply, population migration, and mortality rates possess a bidirectional rather than unidirectional causal structure, it is difficult to capture dynamic interactions solely through the simple establishment of dependent and independent variables.

[0007] Although some statistical techniques, such as the Granger test of causality, were introduced, they had limitations in that they only analyzed the lagged predictive relationships between individual variables and failed to reflect the covariance structure or interdependence of the entire system. Therefore, to comprehensively understand the relationship between urban decline, medical supply, and preventable mortality, a more sophisticated dynamic modeling approach capable of simultaneously considering multiple endogenous variables was required.

[0008] Previous studies have also focused on macroscopic analyses at the single-region or national level, failing to adequately reflect regional heterogeneity. However, even with the same level of medical infrastructure, the effects vary depending on regional socioeconomic conditions, population structure, and migration patterns. Analyses that ignore these differences are highly likely to lead to misjudgments in policy decisions and have limited the establishment of sustainable regional healthcare policies.

[0009] Problems also existed in terms of data processing. Since medical, mortality, and migration statistics are provided by different institutions at varying intervals, time-series inconsistencies and sample imbalances occurred during the integration process. Most studies corrected these issues using simple averaging or missing value imputation, but this resulted in statistical distortion and reduced the predictive power of the models.

[0010] Furthermore, existing studies lacked analytical tools to evaluate the system's response to short-term shocks. For example, when the number of hospital beds temporarily increases or there is a surge in population influx in a specific year, it was difficult to systematically determine how such shocks would affect treatable mortality rates or healthcare supply indicators over the next few years. This served as a major cause of missing the timing for dynamic policy responses.

[0011] Consequently, existing analysis systems have failed to provide a framework capable of interpreting the complex relationships between medical supply, population migration, and mortality rates in a bidirectional and dynamic manner, and lacked a methodological basis to quantitatively verify the effects of policy interventions at the regional level. Therefore, the present invention aims to solve these problems of conventional technology and fundamentally overcome the limitations of conventional technology by providing an analysis system and method that can empirically identify the dynamic interactions between preventable mortality, medical supply, and regional extinction using a Panel Vector Autoregression (PVAR) model. Prior art literature

[0012] Korean Patent Publication No. 10-2709641 (Registration Date: September 20, 2024) The problem to be solved

[0013] The present invention aims to dynamically identify the interdependent and time-lag relationships between treatable mortality rates, medical supply levels, and regional extinction, which are difficult to capture with conventional unidirectional statistical analysis techniques. Existing studies have merely verified the correlation between medical infrastructure and mortality rates or have failed to sufficiently reflect regional heterogeneity and time-series changes. Accordingly, the present invention aims to provide practical evidence for establishing regional medical policies and public health strategies by utilizing a Panel Vector Autoregression (PVAR) approach to comprehensively analyze the interactions between multiple endogenous variables and by visualizing the direction and intensity of influence between variables through Orthogonal Impulse Response Functions (OIRF) and Forwarding Error Variance Decomposition (FEVD). means of solving the problem

[0014] A dynamic interaction analysis system between treatable mortality, medical supply, and local extinction according to one aspect of the present invention for achieving such objectives may comprise: a variable value collection unit that extracts pre-set variable values ​​for a pre-set investigation period and a target group and structures them in the form of a balanced panel; a bidirectional relationship review unit that evaluates the dynamic interdependence between multiple time series of pre-set variable values ​​using a statistical method adopting a PVAR model (panel vector autoregressive model) to derive an estimation result including a bidirectional relationship; and a derived data output unit that processes the estimation result derived through the bidirectional relationship review unit using Orthogonalized Impulse Response Functions (OIRF) and Forecast Error Variance Decomposition (FEVD) and outputs it to an operator.

[0015] In one embodiment of the present invention, in the variable value collection unit, the preset variable values ​​may be the mortality rate due to treatable causes among the population up to 74 years of age (TM_Rate), the number of doctors per 100 people (Physicians), the number of inpatient rooms per 100 people (Inp_Rooms), and the net inflow rate among the population aged 19 to 34 years (Young_Inf).

[0016] In this case, in the variable value collection unit, the ratio of deaths due to treatable causes (TM_Rate) among the population up to age 74 is calculated using the TM rate as a regional health status indicator, based on the cause of death statistics provided by Statistics Korea and the list of preventable and treatable causes of death published by the OECD, and the cause of death statistics may be compiled annually based on death certificates for all domestic deaths to ensure coverage of the entire population.

[0017] In addition, in the variable value collection unit, the treatable causes may include 57 disease groups, including infectious diseases, cardiovascular and cerebrovascular diseases, cancer, endocrine and metabolic diseases, nervous system diseases, circulatory system diseases, respiratory system diseases, and other medical conditions.

[0018] In addition, in the variable value collection unit, the number of inpatient rooms per 100 people (Inp_Rooms) is obtained through regional medical utilization statistics released by the National Health Insurance Service, and since rooms with fewer beds generally accommodate patients requiring more intensive treatment, it may be a variable value extracted by reflecting the overall scale of regional medical resources and considering the difference in treatment intensity per room.

[0019] In addition, in the above variable value collection unit, the net inflow ratio (Young_Inf) among the population aged 19 to 34 is calculated using domestic migration statistics from Statistics Korea and resident registration population statistics from the Ministry of the Interior and Safety, and the resident registration population statistics data from the Ministry of the Interior and Safety is compiled annually based on resident registration records and may be disclosed through the Korea Statistical Information Service.

[0020] In one embodiment of the present invention, in the bidirectional relationship review unit, the PVAR model can evaluate dynamic interdependence between multiple time series by extending a vector autoregression (VAR) model to panel data.

[0021] In addition, in the bidirectional relationship review unit, the PVAR model considers all variables included in the data structured by the variable value collection unit as endogenous variables and can identify the effect that each variable and its lag value have on other variables.

[0022] In addition, in the bidirectional relationship review section, the PVAR model can analyze bidirectional interactions between three or more endogenous variables compared to other existing econometric models.

[0023] In addition, in the bidirectional relationship review section, the PVAR model can control estimation bias caused by unobserved inter-individual heterogeneity by utilizing panel data that includes both individual effects and time-fixed effects.

[0024] In one embodiment of the present invention, the specification of the PVAR model comprises four steps, wherein the four steps include: a step of verifying the static lag condition of the time series within the model by performing a panel unit root test, wherein, since a high lag length of panel data consisting of a relatively short 11-year period can reduce the power of the test due to the loss of degrees of freedom, the maximum lag length is set to 1 and 2 to maintain statistical reliability; a step of determining the optimal lag length, wherein the optimal lag length is determined based on four criteria including the Akaike Information Criterion (AIC), the Schwarz Criterion (SC), the Hannan-Kühn Criterion (HQC), and the Final Prediction Error (FPE); and a step of verifying structural validity by estimating the PVAR model and using Hansen's J statistic (Hansen, 1982), wherein the Generalized Method of Moments (GMM) is adopted for the parameter estimation of the PVAR model, and Hansen's J test is utilized to ensure the exogenous of the instrumental variables required for GMM estimation. The configuration may include a normality test step that verifies the stability and covariance normality of the model by testing whether all eigenvalues ​​of the companion matrix are less than 1 and located within the unit circle.

[0025] In one embodiment of the present invention, in the derived data output unit, the orthogonalized shock response function shows how a shock of one standard deviation to the prediction error of one variable affects the prediction error of the variable itself and other endogenous variables over time, and is generally presented in the form of a plot showing a dynamic response, and when interpreting the results, a positive shock response indicates that the response variable moves in the same direction as the change in the explanatory variable, and a negative shock response may mean the opposite direction.

[0026] In one embodiment of the present invention, in the derived data output unit, the prediction error variance decomposition estimates the proportion of the prediction error variance of each endogenous variable that is attributable to the variable itself and the shock of other variables, and the sum of the variance contributions of all variables to a specific endogenous variable is set to 1, which may mean that the higher the contribution, the greater the influence that shock has on determining the dynamic characteristics of the response variable.

[0027] In one embodiment of the present invention, the derived data output unit further performs a panel Granger causality test (Granger, 1969; Dumitrescu & Hurlin, 2012), and the panel Granger causality test reflects a predictive relationship rather than a true causal relationship, and if a variable X helps improve the prediction of a future value Y, it can be set as a Granger-cause variable Y.

[0028] The present invention may provide a method for analyzing dynamic interactions between treatable mortality, medical supply, and local extinction using the aforementioned dynamic interaction analysis system between treatable mortality, medical supply, and local extinction. A method for analyzing dynamic interactions between treatable mortality, medical supply, and local extinction according to one aspect of the present invention comprises: a variable value collection step of extracting pre-set variable values ​​for a pre-set investigation period and a target group of investigation using a variable value collection unit and structuring them in the form of a balanced panel; and a bidirectional relationship review step of evaluating the dynamic interdependence between multiple time series of pre-set variable values ​​using a statistical method adopting a PVAR model (panel vector autoregressive model) using a bidirectional relationship review unit to derive an estimation result including a bidirectional relationship. The configuration may include a derived data output step that uses a derived data output unit to process the estimation result derived through a bidirectional relationship review unit using Orthogonalized Impulse Response Functions (OIRF) and Forecast Error Variance Decomposition (FEVD), and then outputs it to an operator.

[0029] In one embodiment of the present invention, the bidirectional relationship review step comprises: a step of verifying the positive-negative conditions of the time series within the model by performing a panel unit root test, wherein, since a high lag length of panel data consisting of a relatively short 11-year period can reduce the power of the test due to the loss of degrees of freedom, the maximum lag length is set to 1 and 2 to maintain statistical reliability; a step of determining the optimal lag length, which determines the optimal lag length based on four criteria including the Akaike Information Criterion (AIC), the Schwarz Criterion (SC), the Hannan-Kühn Criterion (HQC), and the Final Prediction Error (FPE); and a step of verifying structural validity by estimating a PVAR model and using Hansen's J statistic (Hansen, 1982), wherein the Generalized Method of Moments (GMM) is adopted for parameter estimation of the PVAR model, and Hansen's J test is utilized to ensure the exogenous of instrumental variables required for GMM estimation. The configuration may include a normality test step that verifies the stability and covariance normality of the model by testing whether all eigenvalues ​​of the companion matrix are less than 1 and located within the unit circle. Effects of the invention

[0030] The dynamic interaction analysis system among preventable mortality, medical supply, and regional extinction according to the present invention provides a statistical framework capable of quantitatively interpreting the complex relationship between regional medical resources and population changes. Through this, the ripple effects of expanding medical personnel and hospital beds on improving public health can be empirically identified, and conversely, the vicious cycle structure in which a shortage of medical supply leads to population outflow and increased mortality can be diagnosed at an early stage. Furthermore, by applying panel data-based PVAR analysis, inter-regional heterogeneity, time-series autocorrelation, and endogeneity issues can be simultaneously controlled, thereby improving policy reliability and prediction accuracy. The visualization results, combining orthogonal impulse response functions and variance decomposition analysis, can be utilized as practical indicators by policymakers to determine priorities for medical and population policies. Consequently, the present invention provides a scientific analytical foundation capable of integrally achieving three public policy goals: alleviating medical supply imbalances, reducing preventable mortality, and preventing regional extinction. Brief explanation of the drawing

[0031] FIG. 1 is a block diagram illustrating a system for analyzing dynamic interactions between treatable mortality, medical supply, and fat loss according to an embodiment of the present invention. FIG. 2 is a flowchart illustrating a method for analyzing dynamic interactions between treatable mortality, medical supply, and fat loss according to an embodiment of the present invention. FIG. 3 is a diagram illustrating the spatial distribution of the incidence rate of leprosy and the supply of medical services in Korea. FIG. 4 is Table 1, illustrating the description and summary statistics of variables. FIG. 5 is a graph illustrating the annual change in national averages by variable from 2012 to 2022. FIG. 6 is Table 2, illustrating the panel unit root test. FIG. 7 is Table 3, illustrating the criteria for selecting the lag order and Hanson's J-test. FIG. 8 is a graph illustrating the stability test of the PVAR model. FIG. 9 is a graph illustrating the OIRF of Model 1. FIG. 10 is Table 4, illustrating the FEVD of Model 1. FIG. 11 is Table 5, illustrating the panel Granger causality test for Model 1. FIG. 12 is the model This is a graph showing the OIRF of 2. Fig. 13 is Table 6 showing the FEVD of Model 2. Fig. 14 is Table 7 showing the panel Granger causality test for Model 2. Fig. 15 is Table A1 showing the PVAR estimation results for Model 1. Fig. 16 is Table A2 showing the PVAR estimation results for Model 2. Specific details for implementing the invention

[0032] Preferred embodiments of the present invention will be described in detail below with reference to the drawings. Prior to this, terms and words used in this specification and claims should not be interpreted as being limited to their ordinary or dictionary meanings, but should be interpreted in a meaning and concept consistent with the technical spirit of the present invention.

[0033] Throughout this specification, when it is stated that one component is located "on" another component, this includes not only cases where one component is in contact with another component, but also cases where another component exists between the two components. Throughout this specification, when it is stated that a part "includes" a component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0034] FIG. 1 shows a block diagram illustrating a system for analyzing dynamic interactions between treatable death, medical supply, and fat loss according to one embodiment of the present invention, and FIG. 2 shows a flowchart illustrating a method for analyzing dynamic interactions between treatable death, medical supply, and fat loss according to one embodiment of the present invention.

[0035] Hereinafter, with reference to FIGS. 1 to 16, each component constituting the dynamic interaction analysis system (100) between treatable death, medical supply, and fat depletion according to the present embodiment will be described in detail.

[0036] Detailed description of the variable value collection unit (110)

[0037] The variable value collection unit (110) serves as the input unit of the analysis system and performs the role of collecting key variable values, such as the treatable mortality rate (TM_Rate), the level of medical supply (Physicians and Inp_Rooms), and the population inflow rate (Young_Inf), from statistical agencies and public databases, and structuring them into a balanced panel. The variable value collection unit (110) is equipped with an interface capable of automatically collecting raw data provided from multiple data sources and incorporates a period-based missing value detection algorithm to ensure time-series consistency of the data. In particular, it is designed to maintain consistency by synchronizing data from different agencies, such as the National Statistical Office, the National Health Insurance Service, and the Ministry of the Interior and Safety, to a single reference year.

[0038] The variable value collection unit (110) includes a missing value correction function to minimize data imbalance issues. For example, if data is missing in a specific year in some regions, the variable value collection unit (110) calculates a reasonable replacement value through the average value of adjacent years, a regression-based predicted value, or an interpolation algorithm. This function plays an important role in preventing distortion of analysis results and ensuring the stability of panel data. In addition, it automatically standardizes differences in data units and scales to facilitate comparison between variables.

[0039] The variable value collection unit (110) clearly defines the calculation process of each variable to increase the transparency and reproducibility of data extraction. For example, TM_Rate is calculated by filtering related disease groups from the Statistics Korea cause of death statistics based on the OECD classification of preventable and treatable causes of death, and converting them into a death rate per 100 people. The number of doctors and the number of hospital beds are each converted into a ratio per 100 people to represent the level of medical supply, and the net population inflow rate is calculated based on the annual migration statistics of the population aged 19 to 34.

[0040] The variable value collection unit (110) stores the extracted data in a standardized panel structure. Each row represents a specific region and year, and each column contains key variables such as TM_Rate, Physicians, Inp_Rooms, and Young_Inf. Through this, observations for all regions and periods are managed in a consistent form and can be directly used as input for the PVAR model.

[0041] Finally, the variable value collection unit (110) incorporates a data logging function to enhance the reliability of the analysis process. All data records the collection time, source, processing stage, and transformation method as metadata, and this metadata is used as reference data for subsequent verification and re-analysis. Additionally, if the user changes the variable definition or period setting, the system automatically re-executes the data collection procedure to immediately provide updated panel data.

[0042] Detailed explanation of the bidirectional relationship review unit (120)

[0043] The bidirectional relationship review unit (120) receives balanced panel data provided by the variable value collection unit (110) and evaluates the dynamic interdependence between treatable mortality, medical supply, and population inflow by applying a panel vector autoregression model (PVAR). This module considers all variables as endogenous variables and estimates the effect of past values ​​of each variable on other variables by considering a time lag structure. In particular, it analyzes not only the effect of changes in medical supply and population inflow on mortality, but also the bidirectional causal structure in which changes in mortality are fed back to medical supply and population migration.

[0044] The bidirectional relationship review unit (120) solves the endogeneity problem by applying the Generalized Method of Moments (GMM) when estimating the model. This ensures consistency in the estimated values ​​even if there is mutual influence between variables, and verifies the exogeneity of instrumental variables through Hansen's J test (Hansen, 1982). This method fundamentally eliminates the estimation bias problem that occurred in existing regression models, making it possible to more accurately identify the impact of changes in medical supply and population structure on mortality rates.

[0045] The bidirectional relationship review unit (120) incorporates a function to verify the stability and validity of the model. It checks the stationarity of each variable using a panel unit root test, and determines the optimal lag length by comprehensively evaluating the AIC, SC, HQC, and FPE criteria. Subsequently, it ensures the stability of the system by verifying whether the eigenvalues ​​of the covariance matrix are located within the unit circle. This stepwise verification procedure increases the reliability of the analysis results and has the effect of preventing statistical errors or overfitting.

[0046] The bidirectional relationship review unit (120) also performs a panel Granger causality test to determine the influence between each variable. This is a prediction-based causality assessment technique that determines whether the past value of one variable improves the prediction of the future value of another variable. For example, if an increase in the number of beds can significantly predict a decrease in future mortality, the number of beds is determined to be the Granger cause of mortality. These analysis results can be used as a practical basis for policymakers to determine the priority of medical resource allocation.

[0047] Finally, the bidirectional relationship review unit (120) includes a function to automatically save and compare model execution results. When iterative analysis is performed by setting different delay lengths or sample intervals, the system automatically compares and organizes the estimated coefficients, test statistics, and goodness-of-fit indicators of each result to present the optimal model specifications. Through this, the user can perform various analysis scenarios based on the same panel data and derive the most realistic and statistically valid interpretation results.

[0048] Detailed explanation of the derived data output unit (130)

[0049] The derived data output unit (130) is an output module that processes and visualizes the estimation results calculated by the bidirectional relationship review unit (120) through an orthogonal impulse response function (OIRF) and a prediction error variance decomposition (FEVD) and provides them to the operator. The orthogonal impulse response function dynamically shows the effect of a 1 standard deviation impulse of a specific variable on the prediction error of that variable and other variables over time, and is visualized in the form of a graph so that the direction and continuity of the interaction between variables can be grasped at a glance.

[0050] The derived data output unit (130) quantitatively calculates the proportion of the variance of the prediction error of each variable that is influenced by other variables through the variance decomposition of the prediction error. For example, if 60% of the prediction error of TM_Rate is explained by changes in the number of beds (Inp_Rooms), it means that the influence of the variable is a policy-important factor. This analysis of the contribution of variance is used as a quantitative basis for determining the priority of resource input when establishing public health policies.

[0051] The derived data output unit (130) provides results in the form of intuitive visual information rather than simple numerical values. It supports various visualization options, such as response function plots, time-variable matrices, and variance ratio graphs, allowing users to easily understand complex dynamic patterns of relationships between variables. Additionally, each analysis result can be exported in PDF, CSV, or API-linked formats, allowing for immediate use in policy reports or academic research.

[0052] The derived data output unit (130) provides a user-customized analysis summary function. It automatically summarizes key statistical indicators of the analysis results (average, maximum effect, duration of response, etc.) and presents the Granger causality results together, thereby enabling a comprehensive understanding of the dynamic structure of the entire system. In particular, by expressing the response intensity that changes along the time axis using a color scale, it allows for a visual understanding of the persistence of influence between variables and the feedback structure.

[0053] Finally, the derived data output unit (130) incorporates a result storage and comparison function to enhance the reproducibility and policy utility of the analysis results. When different model specifications or periods are applied to the same dataset, the system compares the results in parallel and quantifies and presents the differences in impulse response curves for each variable. Through this, users can track how the interaction structure of medical supply, mortality rate, and population inflow changes over time and utilize it to formulate region-specific public health strategies.

[0054] Hereinafter, with reference to FIGS. 1 to 16, each component constituting the dynamic interaction analysis method (S100) between treatable death, medical supply, and fat disappearance according to the present embodiment will be described in detail.

[0055] Detailed explanation of the variable value collection step (S110)

[0056] The variable value collection step (S110) is an input procedure for the analysis system, which involves extracting key indicators such as treatable mortality rates (TM_Rate), medical supply levels (Physicians and Inp_Rooms), and net population inflow rates (Young_Inf) from various public data sources and structuring them into a balanced panel. The variable value collection step (S110) is designed to maintain time-series consistency of the data based on the analysis period (e.g., 2012–2022) and the survey unit (e.g., city, county, district). In particular, it constructs a structured dataset aligned by the same year and same region by integrating and processing data from different agencies, such as the Statistics Korea's mortality statistics, the National Health Insurance Service's regional medical utilization statistics, and the Ministry of the Interior and Safety's resident registration population statistics.

[0057] The variable value collection step (S110) includes data cleaning and standardization processes. Since statistics related to medical supply or population movement may differ in measurement units or base years, differences in scale between variables are corrected through unit conversion (e.g., per person → ratio per 100 people) and log transformation. Additionally, if data for missing years exists, missing values ​​are corrected through time series interpolation or regression-based prediction algorithms to maintain the form of complete panel data for the entire period. This process is a key procedure that ensures the balance of panel data and enhances the reliability of the analysis model.

[0058] In the variable value collection stage (S110), the variable definitions and calculation processes are clearly defined. For example, TM_Rate is calculated by aggregating death data for 57 disease groups, including infectious diseases, circulatory diseases, and neurological diseases, based on the OECD list of preventable and treatable causes of death, and the Physicians variable converts the number of licensed doctors by region into a ratio per 100 people. The Inp_Rooms variable reflects the physical capacity of medical infrastructure by calculating the number of inpatient beds in the region as a ratio per 100 people, and the Young_Inf variable is used as an indicator to measure the risk of local extinction by utilizing net migration statistics of the population aged 19 to 34.

[0059] The variable value collection step (S110) also includes a consistency verification procedure to maintain data quality. By analyzing the yearly fluctuation rate of each variable, abnormal outliers are automatically detected, and data that deviates by more than three standard deviations is classified as a statistical outlier and included in manual review. This process increases the reliability of the analysis by enabling early detection of errors in source data or input mistakes.

[0060] Finally, the variable value collection step (S110) records all collection and transformation processes as metadata, thereby ensuring a reproducible data construction procedure. For example, the data source, collection date, preprocessing method, and missing value handling method are automatically stored as logs and are used as reference material when repeating the same analysis or changing model specifications. Through this, the present invention can create an objective and transparent data-based analysis environment.

[0061] Detailed explanation of the bidirectional relationship review stage (S120)

[0062] The bidirectional relationship review step (S120) is a key step for analyzing the interdependent relationships between treatable mortality rates, medical supply, and population migration by applying a Panel Vector Autoregression (PVAR) model based on the panel data constructed through the variable value collection step (S110). In this step, all variables are considered endogenous variables, and the dynamic influence of past values ​​of each variable on the current values ​​of other variables is simultaneously estimated. For example, it is possible to analyze not only the impact of changes in mortality rates on the expansion of medical supply and population migration, but also the converse impact of medical supply and population migration on changes in mortality rates.

[0063] In the bidirectional relationship review step (S120), the Generalized Method of Moments (GMM) is applied to the model estimation process. This is intended to resolve the endogeneity problem and ensure the consistency of estimates that reflect the mutual influence between variables. During the GMM estimation process, Hansen's J test is used to verify the exogeneity of instrumental variables, thereby ensuring the structural validity of the model. Additionally, data stationarity is ensured through panel unit root testing and static dynamism condition verification (S121), and the dynamic structure of the model is optimized through the optimal lag length determination step (S122).

[0064] The bidirectional relationship review step (S120) is fundamentally different from existing unidirectional regression models in that it can quantitatively identify the feedback structure between medical supply, mortality rates, and population migration. For example, even if an increase in the number of hospital beds contributes to a decrease in mortality rates in the short term, a feedback loop may occur in the long term that increases the population inflow rate, thereby increasing medical demand again. These complex relationships are quantitatively verified through the concurrent estimation structure of the PVAR model.

[0065] In addition, in the bidirectional relationship review stage (S120), the Granger causality test is performed in parallel to further evaluate the predictive relationship between variables. Through this, it is possible to test whether the past value of a specific variable makes a statistically significant contribution to the prediction of the future value of another variable, and to determine which variable acts as a leading indicator for policy purposes. For example, if an increase in mortality significantly predicts a decrease in future population inflow, this empirically supports the linkage between public health issues and regional decline.

[0066] Finally, the bidirectional relationship review step (S120) stores all estimation results in a standardized format and converts them into input values ​​for the Orthogonal Impulse Response Function (OIRF) and Predicted Error Variance Decomposition (FEVD) so that they can be visually analyzed in the derived data output step (S130). Through this, the direction, magnitude, and duration of causal influences between each variable within the model are clearly identified, allowing complex socioeconomic interactions to be directly utilized in policy formulation.

[0067] Detailed explanation of the static-dynamic equivalence condition verification step (S121)

[0068] The static-dynamic condition verification step (S121) is an initial process of the bidirectional relationship review step (S120), and is a step to verify whether time-series variables within the panel data satisfy stationarity. Since the PVAR model operates on the premise of static-dynamic time series, non-stationary variables can lead to incorrect regression results (spurious regression). Therefore, in this step, the stability and statistical validity of the model are ensured by checking whether the mean and variance of each variable remain constant over time.

[0069] In the static equivalence condition verification step (S121), a panel unit root test is applied. The Levin-Lin-Chu (LLC) test, Im-Pesaran-Shin (IPS) test, and Fisher-type test are mainly used, and each test statistically determines whether a unit root exists in an individual time series. In particular, since the analysis period in this invention is relatively short at 11 years, the maximum lag length is set to 1 or 2 to minimize the loss of degrees of freedom and maintain power.

[0070] In this stage, mean-centering and differencing processes are performed on the time series for each region to visually examine the variability and trends of the variables. Variables with long-term trends, such as TM_Rate, are normalized by applying first-order differencing, while the population inflow rate (Young_Inf), which exhibits significant short-term volatility, is log-transformed to ensure variance stability. This preprocessing removes unnecessary noise from the coefficient estimation of the PVAR model and improves the accuracy of the impulse response function analysis.

[0071] The static-dynamic condition verification step (S121) is not limited to simple statistical testing but is directly linked to the subsequent stability evaluation of the model. For example, variables that fail the unit root test are re-tested after differencing, and repeated testing is performed until all endogenous variables satisfy the stationarity condition. This ensures that the covariance structure and the distribution of eigenvalues ​​converge to a stationary state.

[0072] Finally, the static-dynamic condition verification step (S121) maintains analysis transparency by recording the test results as metadata. Test statistics, p-values, differencing status, and the final adopted model for each variable are automatically saved, allowing the user to clearly track the time-series characteristics and normalization process of the variables. Through this procedure, the system of the present invention ensures statistical consistency of the data and maximizes the accuracy and reliability of the bidirectional relationship analysis.

[0073] Detailed explanation of the optimal delay length determination step (S122)

[0074] The optimal lag length determination step (S122) is a step for calculating the optimal lag length to be applied to the PVAR model based on the variables whose stationarity has been secured in the static-dynamic condition verification step (S121). In panel vector autoregression models, variables from past time points are included in the regression equation to express dynamic interactions between variables; however, if the lag length is set too high, a reduction in degrees of freedom and overfitting occur, while conversely, if it is set too short, a problem arises in that the dynamic relationship is not sufficiently reflected. Therefore, this step performs an algorithm that automatically determines the most appropriate lag length by simultaneously considering the explanatory power and simplicity of the model.

[0075] The optimal lag length determination step (S122) evaluates four statistical goodness-of-fit indicators in parallel, including the Akaike Information Criterion (AIC), the Schwarz Criterion (SC), the Hannan-Kühn Criterion (HQC), and the Final Prediction Error (FPE). Each indicator reflects the balance between the sum of squared residuals (SSE) and degrees of freedom differently; for example, AIC emphasizes the predictive power of the model, while SC emphasizes simplicity. The system calculates lag length values ​​from 1 to n for each indicator and determines the lag length that has the minimum value among all criteria as the optimal value.

[0076] In this stage, regional asymmetry is also evaluated by considering the structural characteristics of the panel data. In cases where the sample size is insufficient in some regions, a Common Lag Length is established to ensure degrees of freedom, while a hybrid structure allowing Individual Lag Lengths is applied in data-rich regions. This approach enhances model flexibility and mitigates the unrealistic assumption that all regions possess the same dynamic structure.

[0077] The optimal lag length determination step (S122) has an iterative verification structure. Once the initial candidate lag length is selected, the system verifies the autocorrelation of the residuals, and if the autocorrelation is significant, it increases the lag length by one step and performs re-verification. Through this process, it is possible to sufficiently reflect the interdependence between variables while preventing excessive consumption of degrees of freedom.

[0078] Finally, the optimal lag length determination step (S122) stores the result values ​​in the form of metadata. The information reference values ​​for each lag length, the determined optimal length, and the residual distribution and prediction error structure of the corresponding model are recorded together and are used as reference indicators when reviewing the estimation accuracy of the model in the subsequent structural validity verification step (S123). Through this, this step stabilizes the basic structure of the PVAR model and establishes a reliable foundation for dynamic analysis.

[0079] Detailed explanation of the structural validity verification stage (S123)

[0080] The structural validity verification step (S123) is a step to verify whether the model structure established through the optimal lag length determination step (S122) is statistically consistent and maintains economic significance. Since the PVAR model estimates interactions between multiple endogenous variables, bias may occur in the estimates if the instrumental variables are not set appropriately. To prevent such structural distortion, this step performs an over-identification test using Hansen's J statistic (Hansen, 1982).

[0081] Hansen's J test evaluates the exogeneity of instrumental variables used in GMM (Generalized Method of Moments)-based estimation equations. In other words, it verifies whether the assumption that instrumental variables are independent of the error term is maintained, and if the p-value is above a certain threshold (e.g., p > 0.05), the model structure is deemed valid. This ensures that the PVAR model has a statistically appropriate structure and is free from endogeneity issues.

[0082] The structural validity verification step (S123) goes beyond simple statistical testing to comprehensively evaluate the quantitative stability of the model. The system analyzes the covariance matrix between variables to examine the correlation structure between residual terms and checks for the presence of excessive multicollinearity. Additionally, it checks whether the sign and magnitude of each estimated coefficient are consistent with economic and social interpretations. For example, if a negative coefficient is derived indicating that the mortality rate decreases as medical supply increases, it can be interpreted as strengthening structural validity.

[0083] In this step, confidence intervals for the model's estimation results are also calculated through a bootstrapping procedure. By verifying the distribution characteristics of the coefficients through more than 1,000 resampling processes, the instability of estimation results dependent on a single sample is eliminated. This allows for the securing of robust estimates that reflect the variability of the actual data environment.

[0084] Finally, the structural validity verification step (S123) saves the test results in the form of a summary report and transmits them to the derived data output step (S130). The report includes Hansen's J statistic, p-values, residual autocorrelation test results, and covariance structure analysis results; this information is utilized as a basis for reliability verification and policy interpretation in subsequent analyses. Therefore, this step fundamentally resolves the endogeneity and structural instability issues that occurred in conventional technology, thereby guaranteeing the empirical reliability of the PVAR model.

[0085] Detailed explanation of the normality verification step (S124)

[0086] The normality verification step (S124) is the final step for evaluating the stability and predictability of a model that has passed the structural validity verification step (S123), and is a step for verifying whether the covariance structure of the PVAR model approximates a normal distribution. This is an essential procedure for determining the long-term stability of a dynamic system, and the model maintains a state of stationary equilibrium only if the eigenvalues ​​of the covariance matrix are located inside the unit circle.

[0087] In the normality test step (S124), the distribution characteristics of the residual term are analyzed to determine whether the mean is close to 0 and the variance is constant. A Jarque-Bera test (JB test) is used to determine whether there is abnormal skewness or kurtosis, and if the p-value is greater than or equal to the threshold (0.05), the assumption of normality is considered to be maintained. This process ensures that the probabilistic characteristics of the prediction error are stable, thereby increasing the predictive power and statistical efficiency of the PVAR model.

[0088] The normality test step (S124) also mathematically evaluates the dynamic stability of the system. It tests whether all eigenvalues ​​of the covariance matrix are less than 1, and if some eigenvalues ​​exist outside the unit circle, it re-estimates the model. This process ensures that the model converges stably to a certain level without diverging in response to exogenous shocks.

[0089] In this stage, stability testing is visualized to allow users to intuitively verify the results. The system generates a graph displaying the locations of eigenvalues ​​inside and outside the unit circle, enabling a visual assessment of the long-term equilibrium of the interaction structure between variables. This allows operators to immediately assess the reliability of the analysis results and recognize the need for further readjustment.

[0090] Finally, the normality verification step (S124) stores all verification results in the form of metadata. The residual distribution characteristics, JB test results, eigenvalue spectrum, and stability determination of each variable are recorded, and this is used as the basis for the reliability of the model when performing the Orthogonal Impulse Response Function (OIRF) and Prediction Error Variance Decomposition (FEVD) in the subsequent derived data output step (S130). Therefore, this step solves the unstable prediction problem of conventional technology and completes a high-reliability analysis system that simultaneously guarantees long-term equilibrium and statistical validity.

[0091] Detailed explanation of the derived data output step (S130)

[0092] The derived data output stage (S130) is a stage in which the analysis results, completed up to the bidirectional relationship review stage (S120) and its sub-procedure, the normality verification stage (S124), are processed and visualized to provide to the operator. This stage outputs structured data that allows for the intuitive interpretation of the dynamic relationships between treatable mortality rates (TM_Rate), medical supply (Physicians, Inp_Rooms), and net population inflow rate (Young_Inf) along the time axis, rather than simply listing statistical values. To this end, the system incorporates Orthogonalized Impulse Response Function (OIRF) and Forecast Error Variance Decomposition (FEVD) algorithms, each of which analyzes the direction and intensity of mutual influence between variables.

[0093] In the derived data output step (S130), the Orthogonal Impulse Response Function (OIRF) is calculated first. The OIRF dynamically shows how the prediction errors of a specific variable and other variables change over time when an impulse of one standard deviation occurs to that variable. For example, when the supply of medical services (Physicians) temporarily increases, it shows in the form of a plot how the mortality rate (TM_Rate) and the population inflow rate (Young_Inf) change over the following years. Since this visual representation shows both short-term responses and long-term convergence patterns, the persistence of causal effects between variables and the response speed can be clearly identified.

[0094] Next, Forecast Error Variance Decomposition (FEVD) is performed. FEVD quantitatively calculates the proportion of the forecast error variance of each endogenous variable that is attributed to the impact of the variable itself versus the impact of other variables. For example, if the variation in mortality rates is influenced by medical supply factors by 40% and population inflow factors by 20%, the remaining 40% is interpreted as being attributable to the autovariance of the mortality rate itself. Through this, the relative influence of each factor can be presented numerically, and it is possible to determine which variable improvement contributes most to the stabilization of the overall system from a policy perspective.

[0095] The derived data output step (S130) converts these results into a visual interface. OIRF results are output in the form of a reaction curve graph over time, while FEVD results are output in the form of a stacked bar graph or a pie chart. Each graph displays the time axis (year), response magnitude (±impact), and contribution (%), allowing users to intuitively understand the structure of dynamic interactions between variables. Additionally, the system automatically converts the analysis results into PDF, CSV, image formats, etc., supporting their direct use as policy reports or academic materials.

[0096] Finally, the derived data output step (S130) includes automatic saving and comparison functions to ensure the reusability of verification results and analysis data. When repeated analysis is performed on the same dataset with different periods, variables, or lag lengths, the system presents the differences as standardized indicators by comparing the impulse response graphs, variance contributions, and causality test results of each outcome in parallel. These comparison results can be used by policymakers as a basis for determining the direction of long-term impact resulting from changes in specific variables. Therefore, this step overcomes the limitations of static result interpretation in conventional technology and realizes an advanced analysis and output system capable of interpreting the interdependent structure of medical supply, mortality rates, and population migration in a time-series manner.

[0097] As explained above, the present invention is highly effective in that it overcomes the limitations of analyzing the complex interactions between treatable mortality, medical supply, and regional extinction, which were not resolved in conventional technology. First, existing studies were limited to identifying correlations between individual variables at a single point in time, making it difficult to identify the dynamic impact of changes in regional medical supply on mortality and population migration. By introducing a Panel Vector Autoregression (PVAR) model to simultaneously analyze time-lagged causal relationships among multiple variables, the present invention can empirically identify the chain reaction of changes in each variable on the entire system.

[0098] This invention overcomes the limitations of simple regression models and can reflect the bidirectional relationship between medical resource input and public health outcomes. For example, by considering a structure where the expansion of medical infrastructure acts to reduce mortality rates, while changes in mortality rates in turn influence the allocation of medical resources, it is possible to derive analysis results that closely approximate actual socioeconomic reality. The quantitative analysis of such bidirectional interactions provides policy insights that were impossible with existing unidirectional regression models.

[0099] This invention minimizes the structural bias problem of existing data analysis. Existing studies had low reliability of estimates because they failed to sufficiently control for regional fixed effects or time-series autocorrelation. In this invention, by adopting a panel data-based Generalized Method of Moments (GMM) estimation method, the endogeneity problem is controlled, and more stable and reliable estimation results can be secured by simultaneously considering individual and temporal effects.

[0100] This invention can present the relationships between medical supply, mortality rates, and population migration not merely as simple statistical figures, but as the results of a visual dynamic analysis. By visualizing the direction and intensity of the impact of changes in specific variables on other variables over time in the form of graphs through Orthogonal Impulse Response Functions (OIRF) and Forwarding Error Variance Decomposition (FEVD), the analysis results can be intuitively understood and utilized for policy decision-making. This enables the efficient application of complex econometric models in actual administrative and health policy settings.

[0101] This invention establishes a system capable of evaluating not only quantitative indicators of medical supply but also qualitative fluctuations. By quantitatively measuring the impact of time-series changes in variables—rather than simple indicators such as the number of doctors or hospital beds—on regional mortality rates and population inflow rates, it is possible to empirically assess the issue of qualitative imbalances in medical services. This enables the formulation of medical policies that consider both efficiency and effectiveness, moving away from policies centered solely on resource expansion.

[0102] This invention enhances policy responsiveness to long-term urban decline by simultaneously reflecting regional heterogeneity and time-series changes. By analyzing the self-reinforcing cyclical structure among population inflow rates, changes in medical supply, and mortality rates in each region, it is possible to detect at an early stage when a specific region is exposed to the risk of decline and prevent public health crises through proactive intervention.

[0103] Finally, this invention functions as a practical policy tool that goes beyond mere academic analysis. The structural approach integrating the Panel Granger causality Test enables prediction-based policy formulation and allows for the scientific setting of priorities for complex administrative measures, such as the redeployment of medical personnel, expansion of hospital beds, and policies to attract young people. Therefore, this invention overcomes the limitations of conventional technology and presents a new paradigm for public health management and responses to regional decline by providing quantitative evidence for establishing sustainable regional health and medical policies.

[0104] The above detailed description of the present invention describes only specific embodiments thereof. However, it should be understood that the present invention is not limited to the specific forms mentioned in the detailed description, but rather should be understood to include all variations, equivalents, and substitutions within the spirit and scope of the invention as defined by the appended claims.

[0105] In other words, the present invention is not limited to the specific embodiments and descriptions described above, and any person skilled in the art to which the present invention pertains can make various modifications without departing from the essence of the invention as claimed in the claims, and such modifications fall within the scope of protection of the present invention. Explanation of the symbols

[0106] 100: System for analyzing dynamic interactions between preventable mortality, medical supply, and regional extinction 110: Variable value collection unit 120: Bidirectional relationship review unit 130: Derived data output unit S100: Method for analyzing dynamic interactions between preventable mortality, medical supply, and regional extinction S110: Variable value collection step S120: Bidirectional relationship review step S121: Verification of static-dynamic conditions step S122: Determination of optimal lag length step S123: Structural validity verification step S124: Normality verification step S130: Derived data output step

Claims

Claim 1 A variable value collection unit (110) that extracts pre-set variable values ​​for a pre-set survey period and a survey target group and structures them into a balanced panel form; a bidirectional relationship review unit (120) that evaluates the dynamic interdependence between multiple time series of the pre-set variable values ​​using a statistical method with a PVAR model (panel vector autoregressive model) and derives an estimation result including a bidirectional relationship; The system includes a derived data output unit (130) that processes the estimation result derived through the bidirectional relationship review unit (120) using Orthogonalized Impulse Response Functions (OIRF) and Forecast Error Variance Decomposition (FEVD) and then outputs it; wherein, in the variable value collection unit (110), the preset variable values ​​are the mortality rate due to treatable causes (TM_Rate) among the population up to 74 years of age, the number of doctors per 100 people (Physicians), the number of inpatient rooms per 100 people (Inp_Rooms), and the net inflow rate among the population aged 19 to 34 (Young_Inf); and the mortality rate due to treatable causes (TM_Rate) among the population up to 74 years of age is based on mortality cause statistical data provided by Statistics Korea and prevention data published by the OECD; It is calculated using the list of treatable causes of death, and the above mortality cause statistical data is statistical data compiled on an annual basis covering all domestic deaths, and the above treatable causes include 57 disease groups including infectious diseases, cardiovascular and cerebrovascular diseases, cancer, endocrine and metabolic diseases, neurological diseases, circulatory system diseases, and respiratory system diseases, and the above number of inpatient rooms per 100 people (Inp_Rooms) is calculated using regional medical utilization statistics released by the National Health Insurance Service, and is calculated by considering the scale of regional medical resources and differences in treatment intensity per room,The net inflow ratio (Young_Inf) among the population aged 19 to 34 mentioned above is calculated using domestic migration statistics data from Statistics Korea and resident registration population statistics data from the Ministry of the Interior and Safety, and the resident registration population statistics data from the Ministry of the Interior and Safety is data compiled annually based on resident registration records and disclosed through the Korea Statistical Information Service, and in the bidirectional relationship review unit (120), the PVAR model is a model that evaluates dynamic interdependence between multiple time series by extending a Vector Autoregression (VAR) model to panel data, and the bidirectional relationship review unit (120) considers all variables included in the data structured by the variable value collection unit (110) as endogenous variables and identifies the effect of each variable and its lag value on other variables, analyzes bidirectional interactions between three or more endogenous variables, and removes estimation bias due to unobserved inter-individual heterogeneity using panel data that includes both individual effects and time-fixed effects, and in the derived data output unit (130), the orthogonalized impulse response function is a 1 standard deviation impulse for the prediction error of one variable A dynamic interaction analysis system for treatable mortality, medical supply, and local extinction, characterized by outputting the effect on the prediction error of the variable itself and other endogenous variables over time in the form of a plot, wherein the derived data output unit (130) outputs that if the output of the orthogonalized shock response function is a positive shock response, the response variable moves in the same direction as the change in the explanatory variable, and if the output of the orthogonalized shock response function is a negative shock response, the response variable moves in the opposite direction to the change in the explanatory variable, and outputs the proportion of the prediction error variance of each endogenous variable attributable to the shock of the variable itself and other variables through the prediction error variance decomposition, wherein the sum of the variance contributions of all variables to a specific endogenous variable is set to 1, and outputs the result that the higher the contribution, the greater the influence the shock has on determining the dynamic characteristics of the response variable. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 delete