Rheumatoid arthritis complication prediction system based on asymmetric causal association

Abstract

The invention discloses a rheumatoid arthritis complication prediction system based on asymmetric causal association, and the system comprises the steps: obtaining clinical time series data of a patient, carrying out the preprocessing, carrying out the statistics of the occurrence times of complication disease tags in the clinical data, constructing a disease correlation matrix, calculating the in-degree and out-degree of each complication disease tag pair, and carrying out the calculation of the in-degree and out-degree of each complication disease tag pair. Constructing a complication causal association enhancement matrix; and performing hierarchical clustering on the complication diseases according to the correlation and the causal correlation degree to obtain disease subgroups of complication disease combinations, and performing dynamic prediction on complication risks. According to the method, multi-disease coexistence and dynamic evolution of the RA patient can be reflected more accurately, and a more accurate risk assessment and prediction tool is provided for clinical decision support.

Description

Technical Field

[0001] The present invention relates to the field of medical health information technology, and in particular to a rheumatoid arthritis complication prediction system based on asymmetric causal association. Background Art

[0002] Rheumatoid arthritis (RA) is a chronic, systemic autoimmune disease that is mainly characterized by chronic inflammation and destruction of the joints. However, RA is not limited to the joints, but may also affect multiple systems and organs, leading to a variety of complications. These complications are multiple and organ heterogeneous, seriously affecting the quality of life of patients. Common complications include cardiovascular disease, interstitial lung disease, pulmonary hypertension, etc. These complications not only increase the suffering of patients, but also significantly increase the mortality rate. Therefore, it is crucial to develop effective predictive tools to identify and intervene in these complications early.

[0003] Asymmetric causal association means that in the causal relationship between diseases, the impact of disease A on disease B is not equal and bidirectional, that is, disease A may cause the occurrence of disease B, but the reverse relationship does not hold, or the intensity of the impact between the two is different. For example, in RA patients, pulmonary hypertension and interstitial lung disease often co-morbidity, but pulmonary hypertension is usually caused by interstitial lung disease, not vice versa. Therefore, the causal relationship between diseases is asymmetric, reflecting the differences in directionality and intensity between diseases. This asymmetry reveals the complex interactions between diseases, and the study of asymmetric causal associations helps to identify the dominant factors in specific disease combinations, thereby optimizing early warning and intervention strategies.

[0004] Disease correlation refers to the statistically significant non-random coexistence of two or more health conditions or conditions within a specific population. This means that the frequency of these phenomena occurring together in the same population is higher than the expected level determined by chance alone. Specifically, disease correlation describes the frequency and pattern of occurrence of these diseases in the same population or individual, indicating whether they occur together or there is a certain degree of correlation between their occurrence. Disease correlation can indicate the existence of potential causal relationships and common risk factors. This concept is used to identify potential links between diseases, thereby revealing common pathological pathways, risk factors or shared causal mechanisms. In the field of medical health information, understanding disease associations can help improve diagnosis, predict comorbidities, and design effective treatment and prevention strategies.

[0005] Disease feature similarity refers to the evaluation of the similarity between different diseases by comparing their clinical features, symptoms, signs, pathological features, etc. Specifically, disease feature similarity is to find and measure similar relationships or similar patterns between different diseases by analyzing the multidimensional features of the disease (such as symptoms, clinical data). By analyzing and comparing these features, researchers can find the commonalities between diseases, which helps to understand the occurrence and development mechanism of the disease, identify potential therapeutic targets, and develop new diagnostic tools.

[0006] Causal directionality refers to clearly determining the direction of causal effects in a causal relationship, that is, identifying which variable is the cause and which is the result. It describes which variable's change triggers the change of another variable between a pair of related variables. Causal directionality is not only concerned with whether there is an association (correlation) between variables, but more importantly, it is necessary to clarify the causal direction of this association, that is, which variable is in the position of initiating change in the causal chain and which is the affected object. In the field of medical informatics, disease causal directionality refers to clearly distinguishing the causal relationship between different diseases, that is, which disease is the cause of another disease and which is the result. Specifically, it focuses on whether a change in a disease or health state directly causes the occurrence or progression of another disease. Causal directionality research between diseases helps to reveal the relationship between diseases, understand the mechanism of disease transmission and secondary disease occurrence, and provide guidance for clinical decision-making.

[0007] Dynamic prediction refers to the use of time series data and dynamic models to predict the development trend of the disease, focusing on the timing and variability of the disease process, rather than just predicting based on data at a single point in time. Unlike static cross-sectional predictions, dynamic predictions take into account individual longitudinal data that changes over time, including symptom fluctuations, laboratory test results, treatment responses, etc. In RA patients, as the disease progresses, the patient's health status may change and new complications (such as cardiovascular disease or pulmonary hypertension) may appear. Dynamic prediction can reveal the complex changes in the disease process by modeling time series data. By capturing these dynamic changes, dynamic predictions can more accurately reflect individual disease trajectories, thereby achieving the goal of personalized medicine, especially when dealing with chronic diseases and complex comorbidities.

[0008] Although some progress has been made in the prediction of RA complications in recent years, existing methods still have some limitations:

[0009] 1. Ignoring the complex correlation and synergistic effects between complications

[0010] Existing prediction models usually rely on static cross-sectional data, ignoring the complex interdependence and dynamic associations between complications, and failing to fully capture the causal relationship between diseases and their evolution. This limitation makes it difficult for the model to reveal the interactions and triggering mechanisms between complications, and it is impossible to effectively predict the combined risks of multiple complications. At the same time, existing methods often focus on the prediction of a single complication, failing to fully consider the heterogeneity and synergistic effects between complications, resulting in insufficient ability to identify multiple comorbidity patterns and predict high-risk complication combinations. This limits the accurate prediction of the patient's future disease progression and weakens the implementation of personalized health management and precision intervention.

[0011] 2. Lack of real-time monitoring of dynamic disease progression

[0012] Many existing technologies focus on static data analysis, ignoring dynamic health trends, disease trajectories, and treatment outcomes. Failure to effectively utilize dynamic time series data and track the disease trajectory and comorbidity evolution of rheumatoid arthritis patients in real time leads to insufficient timeliness and accuracy of treatment decisions. Summary of the invention

[0013] The purpose of the present invention is to propose a rheumatoid arthritis complication prediction system based on asymmetric causal association in view of the deficiencies of the prior art. It aims to solve the challenges of current prediction methods in capturing the complex correlation between complications, insufficient prediction ability of comorbidity patterns, and dynamic disease progression prediction. By introducing asymmetric causal association analysis, the system can deeply reveal the triggering and dependency relationships between complications, especially in scenarios where some comorbidities present unidirectional causal effects, such as the relationship between pulmonary hypertension and interstitial lung disease, thereby making up for the shortcomings of existing methods in capturing the correlation of complex diseases. Through time-series dynamic data modeling, the dynamic development of rheumatoid arthritis is accurately tracked, solving the shortcomings of dynamic prediction. By constructing a multi-dimensional cosine similarity matrix and hierarchical clustering analysis, the system can identify high-risk complication combinations and accurately predict comorbidity patterns, further enhancing the ability to predict comorbidity patterns. Combining these innovative technologies, this system significantly improves the accuracy and robustness of rheumatoid arthritis complication prediction, providing strong support for clinical health management and personalized treatment.

[0014] The object of the present invention is achieved through the following technical scheme: a rheumatoid arthritis complication prediction system based on asymmetric causal association, the system comprising:

[0015] Clinical data preprocessing module, used to obtain and preprocess the patient's clinical time series data;

[0016] Complication correlation modeling module, which is used to count the number of occurrences of complication disease labels in clinical data and construct a disease correlation matrix;

[0017] The complication causal association enhancement module is used to calculate the in-degree and out-degree of each complication disease label pair based on the disease correlation matrix, and construct the complication causal association enhancement matrix;

[0018] The comorbidity pattern recognition module is used to hierarchically cluster comorbid diseases according to the correlation and causal association degree based on the complication causal association enhancement matrix to obtain disease subgroups of complication disease combinations;

[0019] The complication prediction module is used to dynamically predict the risk of complications based on disease subgroups of complication disease combinations.

[0020] Furthermore, the clinical data preprocessing module processes and fills the missing values ​​in the time series data by fusing the timestamp, missing mask and time interval.

[0021] Furthermore, in the complication correlation modeling module, the conditional probability of two complication disease labels appearing simultaneously is calculated to construct a disease correlation matrix.

[0022] Furthermore, in the complication causal association enhancement module, for complication disease label pairs, the intermediate labels are enumerated, the contributions of the intermediate labels to the two complication diseases are accumulated, the in-degree matrix is ​​used to measure whether one disease is a prerequisite for the occurrence of another disease, and the out-degree matrix is ​​used to measure the possibility of one disease causing another disease.

[0023] Furthermore, the system also includes a causal analysis auxiliary compensation module, which is used to calculate the similarity between two complication disease labels and combine it with the complication causal association enhancement matrix to obtain a comprehensive association matrix.

[0024] Furthermore, the comprehensive association matrix integrates complication causality and feature similarity information, providing an embedding representation for each complication disease label and capturing the complex associations between complication diseases.

[0025] Furthermore, in the comorbidity pattern recognition module, the average of the causal path weights of the disease in-degree and out-degree of the two complication disease labels is taken as the causal relationship degree of the two complication diseases.

[0026] Furthermore, in the comorbidity pattern recognition module, the similarity between two complication disease clusters is judged based on a comprehensive index combining the correlation and causal relationship between the two complication diseases, and a hierarchical structure is generated by gradually merging the most similar clusters.

[0027] Furthermore, based on the results of the comorbidity pattern recognition module, a hierarchical neural network was designed to predict the development trend of specific disease combinations.

[0028] Furthermore, in the complication prediction module, the dynamic prediction result is the confidence level of each complication disease, reflecting the possibility of rheumatoid arthritis patients suffering from various complications at different time points.

[0029] Beneficial effects of the present invention:

[0030] 1. In-depth understanding of the correlation and synergistic effects between complications: In the clinical practice of RA, patients often have multiple complications, such as cardiovascular disease, lung disease, and osteoporosis. There are complex triggering, dependencies, and causal relationships between these complications, and existing technologies often find it difficult to fully capture these associations. In order to more accurately predict the development trend of the disease and the combination of complications, the present invention proposes a method for deeply capturing the interactions between complication diseases. By introducing asymmetric causal association analysis, it can reveal the complex interactions between different complications and potential disease combination patterns, and comprehensively improve the accuracy and effectiveness of disease prediction. Through this method, doctors can better understand the evolution pattern of RA and its complications, and thus formulate more accurate treatment plans.

[0031] 2. Improve comorbidity pattern recognition and personalized treatment: RA patients often have multiple complications, and the prediction of a single complication cannot meet clinical needs. Traditional technologies ignore the heterogeneity and synergistic effects between complications, resulting in insufficient identification of high-risk complication combinations. The present invention proposes a method for improving comorbidity pattern recognition and personalized patient treatment. Through multi-dimensional causal analysis and hierarchical clustering technology, it can effectively identify and predict synergistic effects and potential high-risk combinations between complications. This not only helps to provide early warning of possible health problems, but also can tailor personalized health management plans for each patient to ensure the effectiveness and safety of treatment.

[0032] 3. Real-time monitoring of disease progression and optimization of treatment decisions: RA is a chronic inflammatory disease whose condition changes over time, so real-time monitoring of the patient's disease status is crucial. Traditional static data evaluation methods are difficult to reflect the true disease dynamics, affecting the timeliness and accuracy of treatment decisions. The present invention proposes a dynamic prediction perspective, combining dynamic time series data and asymmetric causal association analysis, which can monitor the evolution of the disease in real time and provide accurate predictions and interventions based on changes in the patient's health status. Based on dynamic prediction, doctors can detect changes in the disease earlier and adjust treatment strategies in a timely manner to ensure that each patient can obtain the treatment plan that best suits their current conditions.

[0033] 4. The present invention significantly improves the prediction accuracy, robustness and timeliness of rheumatoid arthritis complications, and provides strong support for clinical health management and personalized treatment. This complication prediction system has significant advantages in dealing with the prediction of rheumatoid arthritis complications. Compared with existing technical solutions, it can not only effectively process irregular time series data, but also capture the complex associations between complications, reveal comorbidity patterns, and improve the prediction accuracy of rare complications. In addition, this complication prediction system also has good interpretability and can provide valuable decision support for clinicians. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 It is a schematic diagram of a rheumatoid arthritis complication prediction system based on asymmetric causal association provided by the present invention.

[0036] Figure 2 It is a schematic diagram of causal enhancement.

[0037] Figure 3 This is a schematic diagram of the comorbidity identification module. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific implementation examples described herein are only used to explain the present invention and are not used to limit the present invention.

[0039] like Figure 1 As shown in the figure, the present invention proposes a rheumatoid arthritis complication prediction system based on asymmetric causal association, which introduces dynamic time series data and asymmetric causal relationship modeling between diseases to effectively capture the complexity of rheumatoid arthritis (RA) comorbidities and their dynamic change characteristics. Feature similarity analysis and cosine similarity matrix are used to make up for the missing information in causal relationship modeling, significantly improving the accuracy and robustness of the comorbidity prediction model.

[0040] The system specifically includes a clinical data preprocessing module, a complication correlation modeling module, a complication causal association enhancement module, a causal analysis auxiliary compensation module, a comorbidity pattern recognition module and a complication prediction module. The specific functions of each module are as follows.

[0041] Clinical data preprocessing module, which comprehensively processes and fills missing values ​​in time series data by fusing timestamps, missing masks and time intervals. The integrated time series data is more complete and stable, providing reliable support for clinical decision-making and subsequent data analysis.

[0042] Complication correlation modeling module, when building a rheumatoid arthritis complication prediction system, the correlation between diseases is crucial. As its core pathology, synovitis affects multiple systems throughout the body through inflammatory mediators, such as cardiopulmonary system damage. In addition, although therapeutic drugs (such as immunosuppressants) can control the disease, they may cause side effects in the cardiovascular, liver, kidney and other systems. The complex interactions between these systems make the correlation between complications particularly prominent. Therefore, modeling the correlation between complications and diseases can accurately reflect clinical reality and thus improve system performance.

[0043] The present invention constructs a disease correlation matrix D by counting the number of occurrences of disease labels in the statistics;

[0044]

[0045] Where j and k represent the labels of different complications, respectively. The number of occurrences of complication disease label j, the number of co-occurrences of complication disease label j and complication disease label k, and the conditional probability of the co-occurrence of complication disease label k when complication disease label j appears are calculated. .

[0046] Modeling the correlation between complication diseases can capture the mutual influence and dependency between different diseases, more realistically reflect the complex pathological and treatment mechanisms in clinical practice, and thus significantly improve the predictive performance of the system.

[0047] Complication causal relationship enhancement module, because there are often complex causal relationships between different complications, the disease relationship shows directional characteristics. For example, there may be a frequent co-occurrence relationship between "pulmonary hypertension" and "interstitial lung disease", but this relationship is often unidirectional and asymmetric. "Pulmonary hypertension" may often be accompanied by "interstitial lung disease", but the opposite is not necessarily true. Figure 2 As shown, in order to accurately capture this asymmetric disease label relationship, the present invention constructs a complication causal relationship enhancement matrix based on the proposed disease correlation matrix, and uses the matrix representation of in-degree and out-degree to reveal the mutual influence between complication diseases and potential comorbidity patterns, thereby deepening the understanding of specific complication disease combinations.

[0048]

[0049]

[0050] For each complication disease label pair , enumerate the intermediate labels u, calculate and , respectively, the normalized denominators are and , v refers to all labels in the label set, and the contribution of u to diseases j and k is accumulated. Measures whether disease k is a prerequisite for the occurrence of disease j, the out-degree matrix A measure of the likelihood that disease j will cause disease k.

[0051] The auxiliary compensation module of causal analysis, relying only on the in-degree and out-degree matrices, may not be able to fully capture the combination of complications that are not directly causally related, especially the complication associations that show strong statistical correlation but lack significant causal relationships. Therefore, the cosine similarity matrix is ​​introduced as a similarity measurement method independent of the statistical characteristics of the data. By quantifying the similarity of complication diseases in a multidimensional feature space, cosine similarity can identify complication relationships that have similar characteristics but may not be obvious in statistical frequency. It provides auxiliary information when the causal relationship is not sufficient to explain, thereby effectively making up for the blind spots in directional analysis.

[0052]

[0053] in, They represent the feature vector representations of complication disease label j and complication disease label k respectively.

[0054] The final comprehensive correlation matrix M that integrates the causal relationship of complications and feature similarity information is:

[0055]

[0056] It is a weight matrix that linearly transforms the input features, which determines how to adjust the importance of the input features and how to project the input data into a new space or dimension. It is a bias term used to adjust the results after linear transformation to help the model capture more complexity and nonlinear relationships.

[0057] The combined association matrix M provides a final embedding representation for each label.

[0058] The comprehensive association matrix M comprehensively captures the complex associations between RA complications by integrating causal relationships and feature similarity information, significantly improving the performance of the prediction system and supporting more accurate clinical decision-making and personalized treatment planning.

[0059] Comorbidity pattern recognition module,In order to further support the complication prediction system to achieve complication prediction and deeply explore disease association patterns, the present invention designs a disease relationship layered auxiliary task module to identify comorbidity patterns and reveal high-risk complication combinations.

[0060] like Figure 3 As shown in the figure, the comorbidity pattern recognition module first quantifies the correlation and causal relationship between diseases, and clusters complications with high correlation and high causal relationship into the same group. This step can not only reveal the comorbidity characteristics between complications, but also further perform cluster analysis on patients, help identify comorbidity patterns and their potential mechanisms within subgroups, and reveal high-risk complication combinations.

[0061] The present invention defines a comprehensive index To describe the correlation and causal relationship between diseases j and k:

[0062]

[0063] in, and Respectively represent the causal path weights from the disease in-degree and out-degree, is a parameter between 0 and 1 that adjusts the relative importance between correlation and causation.

[0064] In hierarchical clustering, a pair of clusters whose elements are most similar are merged:

[0065]

[0066] in, The clusters representing complication disease labels j and k, respectively. A hierarchical structure was generated by stepwise merging the most similar clusters, revealing different subgroups of complication combination characteristics and potential high-risk complication combinations.

[0067] Based on the clustering results of the comorbidity pattern recognition module, the present invention further designed a hierarchical neural network that focuses on predicting the development trend of specific disease groups. This architecture not only helps to understand the interaction between diseases, but also provides a new perspective for the precise grouping of patients, making it possible to identify high-risk complication combinations at an early stage.

[0068] The complication prediction module, based on the above modules, performs multi-label prediction for seven common complications of RA patients, including interstitial lung disease, mental illness, pulmonary hypertension, infection, cardiovascular disease, skin ulcers and bone marrow suppression.

[0069] The present invention has achieved good prediction results on the data of rheumatoid arthritis direct reporting projects from 274 rheumatology centers in 31 provinces in China. The prediction accuracy rates of 7 complications, including interstitial lung disease, mental illness, pulmonary hypertension, infection, cardiovascular disease, skin ulcer and bone marrow suppression, are 0.941, 0.976, 0.996, 0.998, 0.944, 0.992 and 0.997 respectively, and the recall rates are 0.763, 0.746, 0.728, 0.668, 0.805, 0.747 and 0.659 respectively.

[0070] The system of the present invention can dynamically predict the risk of complications of RA patients and display the prediction results in real time. The displayed results are the confidence level of each complication, which can intuitively reflect the possibility of RA patients suffering from various complications at different time points. Through this prediction system, clinicians can make more accurate personalized treatment decisions based on real-time updated risk assessments, thereby optimizing patient management and improving the effectiveness of clinical interventions.

[0071] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A rheumatoid arthritis complication prediction system based on asymmetric causal association, characterized in that: The system includes: Clinical data preprocessing module, used to obtain and preprocess the patient's clinical time series data; Complication correlation modeling module, which is used to count the number of occurrences of complication disease labels in clinical data and construct a disease correlation matrix; The complication causal association enhancement module is used to calculate the in-degree and out-degree of each complication disease label pair based on the disease correlation matrix, and construct the complication causal association enhancement matrix; The comorbidity pattern recognition module is used to hierarchically cluster comorbid diseases according to the correlation and causal association degree based on the complication causal association enhancement matrix to obtain disease subgroups of complication disease combinations; The complication prediction module is used to dynamically predict the risk of complications based on disease subgroups of complication disease combinations.

2. A rheumatoid arthritis complication prediction system based on asymmetric causal association according to claim 1, characterized in that: The clinical data preprocessing module processes and fills the missing values ​​in the time series data by fusing the timestamp, missing mask and time interval.

3. A rheumatoid arthritis complication prediction system based on asymmetric causal association according to claim 1, characterized in that: In the complication correlation modeling module, the conditional probability of two complication disease labels appearing simultaneously is calculated to construct a disease correlation matrix.

4. A rheumatoid arthritis complication prediction system based on asymmetric causal association according to claim 1, characterized in that: In the complication causal association enhancement module, for complication disease label pairs, the intermediate labels are enumerated, and the contributions of the intermediate labels to the two complication diseases are accumulated. The in-degree matrix is ​​used to measure whether one disease is a prerequisite for the occurrence of another disease, and the out-degree matrix is ​​used to measure the possibility of one disease causing another disease.

5. The rheumatoid arthritis complication prediction system based on asymmetric causal association according to claim 1, characterized in that: The system also includes a causal analysis auxiliary compensation module, which is used to calculate the similarity between two complication disease labels and combine it with the complication causal association enhancement matrix to obtain a comprehensive association matrix.

6. A rheumatoid arthritis complication prediction system based on asymmetric causal association according to claim 5, characterized in that: The comprehensive association matrix fuses complication causality and feature similarity information, providing an embedding representation for each complication disease label and capturing the complex associations between complication diseases.

7. The rheumatoid arthritis complication prediction system based on asymmetric causal association according to claim 1, characterized in that: In the comorbidity pattern recognition module, the average of the causal path weights of the disease in-degree and out-degree of the two complication disease labels is taken as the causal relationship degree of the two complication diseases.

8. The rheumatoid arthritis complication prediction system based on asymmetric causal association according to claim 1, characterized in that: In the comorbidity pattern recognition module, the similarity of two complication disease clusters is judged based on a comprehensive index combining the correlation and causal relationship between the two complication diseases, and a hierarchical structure is generated by gradually merging the most similar clusters.

9. The rheumatoid arthritis complication prediction system based on asymmetric causal association according to claim 1, characterized in that: Based on the results of the comorbidity pattern recognition module, a hierarchical neural network was designed to predict the development trend of specific disease combinations.

10. The rheumatoid arthritis complication prediction system based on asymmetric causal association according to claim 1, characterized in that: In the complication prediction module, the dynamic prediction result is the confidence level of each complication disease, reflecting the possibility of rheumatoid arthritis patients suffering from various complications at different time points.

Images