Dynamic risk prediction system for breast cancer-related lymphedema

Abstract

The invention discloses a breast cancer related lymphedema dynamic risk prediction system, and relates to the technical field of breast cancer treatment, and the system comprises a data collection module which continuously collects the social demographic characteristics, physiological data, disease and treatment information and life behavior data of a breast cancer patient based on the social ecosystem theory; the data preprocessing module is used for cleaning and processing original data; the variable screening module is used for selecting important static and dynamic variables and constructing a dynamic prediction model based on evidence, expert opinions and a statistical analysis method; and the dynamic combined prediction model module is used for dynamically predicting the lymphedema occurrence risk of the patient through a longitudinal sub-model and a survival sub-model by adopting a combined survival analysis model. The accurate, personalized and real-time breast cancer related lymphedema risk assessment method is provided through multi-dimensional data acquisition, fine variable screening and optimization, construction of a dynamic prediction model, expert feedback and continuous adjustment, the health management level of breast cancer patients is remarkably improved, and the incidence rate of lymphedema is reduced.

Description

Technical Field

[0001] This invention relates to the field of breast cancer treatment technology, specifically to a dynamic risk prediction system for breast cancer-related lymphedema. Background Technology

[0002] Breast cancer is one of the most common cancers among women worldwide, and with advancements in treatment technology, its survival rate has significantly improved. However, complications such as lymphedema often occur during breast cancer treatment, severely impacting patients' quality of life. Lymphedema is a type of interstitial edema caused by obstructed lymphatic drainage. It is particularly prevalent during breast cancer treatment due to the side effects of lymph node dissection, radiotherapy, or chemotherapy.

[0003] Currently, risk assessment of breast cancer-related lymphedema relies on static factors, such as the patient's clinical characteristics and treatment plan. However, the health status of breast cancer patients changes dynamically during treatment, affecting the risk of lymphedema. Therefore, existing static risk assessment methods cannot accurately assess the dynamic risk of patients in real time, and a new method is urgently needed to predict the risk of lymphedema in patients in real time over a period of time. Summary of the Invention

[0004] To address the aforementioned technical problems, a dynamic risk prediction system for breast cancer-related lymphedema is provided. This technical solution resolves the problems described above.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A dynamic risk prediction system for breast cancer-related lymphedema includes:

[0007] The data acquisition module is used to dynamically collect sociodemographic characteristics, physiological data, disease and treatment information, lifestyle and behavioral data, and psychosocial data of breast cancer patients.

[0008] The data preprocessing module is electrically connected to the data acquisition module. The data preprocessing module is used to clean and process the raw data, including outlier detection and processing, data merging and variable transformation, and missing value imputation, to ensure the accuracy and integrity of the data.

[0009] The variable screening module is electrically connected to the data preprocessing module. The variable screening module screens important static and dynamic variables based on evidence-based evidence, expert opinions and statistical analysis methods, and constructs a dynamic prediction model.

[0010] The dynamic joint prediction model module is electrically connected with the variable screening module, adopts a joint survival analysis model, and dynamically predicts the risk of lymphedema of the patient through a longitudinal sub-model and a survival sub-model.

[0011] Preferably, the data acquisition module specifically comprises:

[0012] The socio-demographic data acquisition unit is used to acquire the socio-demographic information of the patient, including age, education level, marital status, economic status, and medical insurance type;

[0013] The physiological data acquisition unit is used to acquire the lymphedema-related physiological data of the patient in real time, including body mass index and bilateral upper limb circumference;

[0014] The clinical data acquisition unit is used to collect the clinical treatment information of the patient in real time, including the maximum diameter of the tumor, the type of breast surgery, the type of axillary surgery, the number of cleaned lymph nodes, the number of positive lymph nodes, the postoperative complication situation, the received chemotherapy scheme, and the postoperative radiotherapy information;

[0015] The life behavior habit data acquisition unit is used to monitor the self-management behavior and life behavior habit of the patient related to lymphedema, including lymphedema self-management behavior, lymphedema risk and protection behavior, and exercise behavior;

[0016] The psychosocial data acquisition unit collects the psychosocial variables related to the lymphedema of the patient, including disease perception, anxiety, depression, self-efficacy, and medical social support;

[0017] The data preprocessing module cleanses and processes different types of data through intelligent algorithms to ensure the quality and consistency of the original data, especially through mean interpolation and Z-score outlier detection technology, so that the data after processing can be used for more accurate statistical analysis and model construction, reducing the influence of data missing and errors on the prediction results.

[0018] Preferably, the data preprocessing module specifically comprises:

[0019] The repeated data checking and deduplication unit checks whether there are repeated records in the data set, deletes the repeated data records, and ensures that each data point is unique;

[0020] The outlier detection and processing unit uses Z-score detection to identify outliers in the data, and replaces the outliers with the mean value of the variable;

[0021] The data normalization and standardization unit normalizes the data, converts features of different scales to the same range, and converts the data to a standard normal distribution;

[0022] Missing value processing unit: fill in missing values based on mean imputation, if the proportion of missing values > 30%, select to merge with adjacent data points or delete the data points;

[0023] Data merging and linking unit: merge data from different data sources and different collection time points to ensure data integrity;

[0024] The module combines statistical analysis methods and expert opinions, adopts multiple variable screening techniques such as Pearson correlation coefficient and statistical significance test, screens out features that have a significant impact on breast cancer-related lymphedema risk from a large number of potential variables, and distinguishes between dynamic and static variables to ensure that the features on which the final model is based have the most predictive power.

[0025] Preferably, the variable screening module specifically includes:

[0026] Feature selection algorithm unit: statistically significant test is used to evaluate each variable individually and select variables that have a significant impact on the target variable;

[0027] Static and dynamic variable identification unit: select features that do not change over time as static features, including user's gender, age, education level and region, and select features that change over time as time series data, including user's body mass index, self-management behavior and exercise changes;

[0028] Variable importance verification and cross-validation unit: use cross-validation techniques to evaluate the performance of selected features in the prediction model, verify the effectiveness of feature selection, build a preliminary model to evaluate the contribution of different features to the prediction result, and further screen and verify the optimal feature set;

[0029] By combining longitudinal sub-models with survival sub-models and using joint survival analysis methods, the system can dynamically predict the risk of lymphedema, and the design of the joint model breaks through the limitations of traditional single survival analysis models, allowing more comprehensive consideration of patients' various clinical information and dynamic data over time, thereby providing personalized risk assessment.

[0030] Preferably, the use of Pearson correlation coefficient to evaluate the correlation between variables and remove variables with low correlation with the target variable specifically includes:

[0031] Wherein, the Pearson correlation coefficient formula is:

[0032]

[0033] In the formula, r is the Pearson correlation coefficient, x i y i are the data points of two variables, is the mean of variables x and y.

[0034] Preferably, the evaluation of each variable by statistical significance test is performed individually, and the selection of variables that have a significant impact on the target variable specifically includes:

[0035] wherein the statistical significance test formula is:

[0036]

[0037] wherein, is the mean of two samples, is the variance of two samples, n1n2 is the sample size of two samples, t is the significance, if the t value is less than 0.05, the feature has a significant impact on the target variable, and the feature is retained.

[0038] Preferably, the dynamic joint prediction model module specifically includes:

[0039] A longitudinal sub-model construction unit: based on the time series data during patient follow-up, a longitudinal sub-model is established to capture the dynamic changes of the patient's lymphedema-related conditions at different time points, including the patient's body mass index, disease perception, anxiety and depression, self-efficacy, medical social support, and changes in self-management behavior. The longitudinal sub-model is trained, and through analysis of patient follow-up data, dynamic change factors affecting the occurrence of lymphedema are identified.

[0040] A survival sub-model construction unit: a Cox proportional hazards model is used to analyze the risk of patients developing lymphedema, and based on the output of the Cox proportional hazards model, the risk of each patient developing lymphedema in the future is obtained.

[0041] The longitudinal sub-model can capture the dynamic changes in health status through patient follow-up data, providing time series data analysis capabilities, while the survival sub-model uses the Cox proportional hazards model to predict the risk of developing lymphedema. By combining the two, a more comprehensive and efficient risk prediction system can be formed.

[0042] Preferably, the dynamic joint prediction model module further includes:

[0043] A joint model construction unit: combining the longitudinal sub-model and the survival sub-model, a joint survival analysis method is used to integrate information from the two sub-models, considering the mutual relationship between longitudinal data and survival data.

[0044] A model evaluation and optimization unit: the C-index consistency index, Brier calibration score, sensitivity, specificity, positive predictive value, negative predictive value, balanced accuracy, and clinical decision curve are used to evaluate the prediction performance and clinical utility of the joint survival analysis model.

[0045] Dynamic prediction and risk assessment unit: based on the trained joint survival analysis model, dynamic prediction is carried out, the risk of lymphedema of the patient is updated in real time, the patient is given personalized risk assessment, and the survival curve of the patient is generated;

[0046] wherein the dynamic prediction formula is:

[0047]

[0048] wherein, is the instantaneous risk of the event occurring at time t under the given covariate X t , h0(t) is the baseline risk function, exp is the exponential function, and the linear combination of covariates and regression coefficients is converted into a risk multiplication factor, β1, β2, β p are regression coefficients, X t1 , X t2 , X tp is the value of the specific covariate at time t;

[0049] Expert feedback and adjustment unit: according to the feedback of doctors and field experts, the model is further adjusted, the structure, input features and training process of the model are continuously optimized by combining the feedback of experts and the actual performance of the model;

[0050] Visualization and report generation unit: generate a visual report of the patient's lymphedema risk, including survival curve and risk probability curve chart;

[0051] Model evaluation and optimization: the use of C-index consistency index effectively measures the prediction accuracy of the model, and the model is continuously adjusted and optimized through expert feedback to ensure that the prediction results can be updated in real time and accurately predict the risk of the patient. This process enables the system to continuously improve the prediction performance in actual use.

[0052] Preferably, the longitudinal sub-model construction unit specifically comprises: wherein the longitudinal sub-model formula is:

[0053]

[0054] wherein y i (t) represents the longitudinal result observed at time point t of the i th research object, ε i (t) is the measurement error, m i (t) represents the true, unobserved longitudinal result, affected by measurement error, β, b i are fixed effect and random effect coefficients, respectively, are the row vectors of the fixed effect and random effect matrices, respectively, b i~N(0,D) represents hypothesis b i It follows a multivariate normal distribution with mean 0 and variance D, ε i (t)~N(0,σ 2 ) indicates the assumption ε i (t) follows a mean of 0 and a variance of σ. 2 The multivariate normal distribution.

[0055] Preferably, the survival sub-model construction unit specifically includes:

[0056] The Cox proportional hazards model formula is as follows:

[0057]

[0058] In the formula, h i (t|M i (t),ω i Let represent the risk function of individual i at time t. This section presents the instantaneous risk function, which describes the instantaneous probability of an event occurring near time t for individual i under given conditions. This represents the instantaneous risk rate as the time interval shrinks infinitely. dt represents the length of the time interval, and the limit definition of instantaneous probability is achieved by dt→0. This represents the event time of individual i when the following conditions are met. The conditional probability that the value falls within the interval (t+dt). This indicates that the individual has survived to time t (i.e., before the event has occurred). i (t|M i (t),ω i )=h0(t)exp{γ T ω i +αm i The formula (t)}, t>0 is an extension of the Cox proportional hazards model in a joint model, used to analyze the association between longitudinal data and survival data, with the aim of quantifying the longitudinal process m. i (t) How does it affect the risk of the outcome event? Traditional Cox models only include the baseline covariate ω. i This model uses αm i (t) introduces longitudinal information that changes over time.

[0059] This module uses dynamic prediction formulas to update the risk of lymphedema in real time as a patient's health data changes. Based on the current changes in covariates and regression coefficients, it makes accurate risk estimates for the future, providing strong support for precise risk control and individualized treatment.

[0060] Compared with the prior art, the application has the beneficial effects that:

[0061] The application provides an accurate, personalized and real-time breast cancer related lymphedema risk assessment method by multi-dimensional data collection, dynamic prediction model construction, fine variable screening and optimization, and expert feedback and continuous adjustment, significantly improves the health management level of breast cancer patients, and provides a powerful support tool for the clinic, helps to optimize the treatment plan and life management of patients, reduces the incidence of lymphedema, and improves the quality of life of patients. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 The system framework of the application is shown in the figure;

[0063] Figure 2 The internal system framework of the dynamic joint prediction model module in the application is shown in the figure. DETAILED DESCRIPTION

[0064] The following description is used to disclose the application so that those skilled in the art can implement the application. The preferred embodiments in the following description are only as examples, and other obvious modifications can be made by those skilled in the art.

[0065] Referring to Figure 1 The breast cancer related lymphedema dynamic risk prediction system shown in the figure comprises:

[0066] A data collection module is used to dynamically collect the social demographic characteristics, physiological data, disease and treatment information, life behavior habit data and psychosocial data of breast cancer patients;

[0067] A data preprocessing module is electrically connected with the data collection module, and the data preprocessing module is used to clean and process the original data, including abnormal value inspection and processing, data merging and variable conversion, and missing value interpolation, to ensure the accuracy and integrity of the data;

[0068] A variable screening module is electrically connected with the data preprocessing module, and the variable screening module screens important static and dynamic variables based on evidence-based evidence, expert opinions and statistical analysis methods, and constructs a dynamic prediction model;

[0069] A dynamic joint prediction model module is electrically connected with the variable screening module, and the dynamic joint prediction model module adopts a joint survival analysis model to dynamically predict the risk of lymphedema of the patient through a longitudinal sub-model and a survival sub-model.

[0070] The data collection module specifically comprises:

[0071] Sociodemographic data collection unit: for collecting patients' sociodemographic information, including age, education level, marital status, economic status, and medical insurance type;

[0072] Physiological data collection unit: for collecting patients' lymphedema-related physiological data in real time, including body mass index, bilateral upper limb circumference;

[0073] Clinical data collection unit: for collecting patients' clinical treatment information in real time, including BMI, maximum tumor diameter, number of lymph node dissection, received chemotherapy regimen, and postoperative radiotherapy information;

[0074] Lifestyle habit data collection unit: for monitoring patients' self-management behaviors and lifestyle habits related to lymphedema, including lymphedema self-management behaviors, lymphedema risk and protective behaviors, and exercise behaviors;

[0075] Psychosocial data collection unit: for collecting patients' psychosocial variables related to lymphedema, including disease perception, anxiety, depression, self-efficacy, and medical social support.

[0076] The data preprocessing module specifically includes:

[0077] Duplicate data checking and deduplication unit: checks whether there are duplicate records in the data set, deletes duplicate data records, and ensures that each data point is unique;

[0078] Outlier detection and processing unit: uses Z-score detection to identify outliers in the data, and replaces outliers with the mean value of the variable;

[0079] Data normalization and standardization unit: normalizes the data, converts features of different scales to the same range, and converts the data to a standard normal distribution;

[0080] Missing value processing unit: based on mean imputation, missing values are filled, if the proportion of missing values is >30%, the data point is merged with the adjacent data point or deleted;

[0081] Data merging and connection unit: merges data from different data sources and different collection time points to ensure data integrity.

[0082] The variable screening module specifically includes:

[0083] Feature selection algorithm unit: each variable is evaluated individually through statistical significance test, and variables that have a significant impact on the target variable are selected;

[0084] Static and dynamic variable identification unit: Select features that do not change over time as static features, including users' gender, age, education level, and region; select features that change over time as time series data, including users' body mass index, self-management behavior, and changes in exercise status.

[0085] Variable Importance Validation and Cross-Validation Unit: Using cross-validation techniques, the unit evaluates the performance of selected features in the prediction model, verifies the effectiveness of feature selection, evaluates the contribution of different features to the prediction results by building a preliminary model, and further filters and validates the optimal feature set.

[0086] Using the Pearson correlation coefficient to assess the correlation between variables, variables with low correlation to the target variable are removed, specifically including:

[0087] The formula for the Pearson correlation coefficient is as follows:

[0088]

[0089] In the formula, r is the Pearson correlation coefficient, and x i y i These are data points for two variables. It is the mean of variables x and y.

[0090] Each variable was evaluated individually using statistical significance testing. Variables that had a significant impact on the target variable were specifically selected as follows:

[0091] The formula for the statistical significance test is as follows:

[0092]

[0093] In the formula, It is the mean of two samples. is the variance of the two samples, n1 and n2 are the sample sizes of the two samples, and t is the significance. If the t value is less than 0.05, then the feature has a significant impact on the target variable and the feature is retained.

[0094] Reference Figure 2 As shown, the dynamic joint prediction model module specifically includes:

[0095] Longitudinal sub-model building unit: Based on time series data during patient follow-up, a longitudinal sub-model is built to capture the dynamic changes of patients' lymphedema-related conditions at different time points, including changes in patients' body mass index, disease perception, anxiety and depression, self-efficacy, medical and social support, and self-management behavior. The longitudinal sub-model is trained, and through the analysis of patient follow-up data, dynamic factors affecting the occurrence of lymphedema are identified.

[0096] Survival sub-model construction unit: Cox proportional hazards model is used to analyze the risk of patients developing lymphedema, and based on the output of the Cox proportional hazards model, the risk of each patient developing lymphedema in the future is obtained.

[0097] The dynamic joint prediction model module further comprises:

[0098] Joint model construction unit: combining the longitudinal sub-model and the survival sub-model, using joint survival analysis method, integrating information from the two sub-models, considering the mutual relationship of longitudinal data and survival data;

[0099] Model evaluation and optimization unit: using C-index consistency index, Brier calibration score, sensitivity, specificity, positive predictive value, negative predictive value, balanced accuracy and clinical decision curve to evaluate the prediction performance and clinical practicability of the joint survival analysis model;

[0100] Dynamic prediction and risk assessment unit: based on the trained joint survival analysis model, dynamic prediction is carried out, the risk of developing lymphedema of patients is updated in real time, personalized risk assessment is carried out for patients, and the survival curve of patients is generated;

[0101] Wherein, the dynamic prediction formula is:

[0102]

[0103] In the formula, is the instantaneous risk of the event occurring at time t under the given covariate X t , h0(t) is the baseline risk function, exp is the exponential function, and the linear combination of covariates and regression coefficients is converted into the product factor of risk, β1, β2, β p are regression coefficients, X t1 , X t2 , X tp is the value of a specific covariate at time t;

[0104] Expert feedback and adjustment unit: according to the feedback of doctors and field experts, the model is further adjusted, combining the feedback of experts and the actual performance of the model, the structure, input features and training process of the model are continuously optimized;

[0105] Visualization and report generation unit: generate visual report of patient lymphedema risk, including survival curve and risk probability curve chart.

[0106] The longitudinal sub-model construction unit specifically comprises:

[0107]

[0108] In the formula, yi (t) denotes the longitudinal outcome observed for the i-th study subject at time point t, ε i (t) is the measurement error, m i (t) denotes the true, unobserved longitudinal outcome, affected by measurement error, β, b i are the fixed effects and random effects coefficients, respectively, are the row vectors of the fixed effects and random effects matrices, respectively, b i ~N(0, D) denotes the assumption that b i follows a multivariate normal distribution with mean 0 and variance D, ε i (t) ~ N(0, σ 2 ) denotes the assumption that ε i (t) follows a multivariate normal distribution with mean 0 and variance σ 2 .

[0109] The survival submodel construction unit specifically comprises:

[0110] The Cox proportional hazards model formula is:

[0111]

[0112] In the formula, h i (t | M i (t), ω i ) denotes the risk function of individual i at time t, This part is the instantaneous risk function, which describes the instantaneous probability of individual i occurring an event near time t under given conditions. represents the instantaneous risk rate when the time interval is infinitely small. dt represents the length of the time interval, and the limit definition of the instantaneous probability is achieved by dt→0. represents the conditional probability that the event time of individual i falls within the interval (t+dt) when the following conditions are met. represents the conditional probability that the event time of individual i falls within the interval (t+dt) when the following conditions are met. represents that individual i survives to time t (i.e., has not yet occurred an event). h i (t | M i (t), ω i ) = h0(t)exp{γ T ω i + αm i (t)}, t > 0 This formula is an extended form of the Cox proportional hazards model in the joint model, which is used to analyze the association between longitudinal data and survival data, and the purpose is to quantify how the longitudinal process m i (t) affects the risk of the outcome event. The traditional Cox model only contains the baseline covariate ω i , while this model includes αmi (t) longitudinal information over time is introduced.

[0113] In summary, the advantages of the present application are:

[0114] By collecting various data sources such as sociodemographic data, physiological data, clinical data, lifestyle habits and self-management behavior data, and psychosocial data, the health status of breast cancer patients is comprehensively evaluated, thereby providing more accurate data support for risk prediction;

[0115] By establishing the combination of longitudinal sub-models and survival sub-models, dynamic prediction is realized, which can update the risk assessment of patients in real time, and consider the changes in patients' health and external interventions;

[0116] The system screens important variables by combining evidence-based evidence, expert opinions and statistical analysis, ensuring the scientificity, effectiveness and accuracy of the prediction model, uses cross-validation and statistical significance test to further optimize the model features, and improves the reliability of the prediction results;

[0117] The system can provide personalized risk assessment for each patient, generate survival curves and risk probability curves, provide strong decision support for doctors, help patients adjust treatment plans and lifestyle habits, and reduce the risk of lymphedema;

[0118] By regularly collecting feedback from experts, the structure and input features of the prediction model are continuously adjusted and optimized, so that the system can better adapt to clinical practice and improve the accuracy and clinical value of risk prediction;

[0119] The system can generate a visual report of the risk of lymphedema, allowing patients and doctors to intuitively understand the risk status of the patient, and provide a scientific basis for subsequent treatment decisions.

[0120] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A breast cancer related lymphedema dynamic risk prediction system, comprising: The application relates to a breast cancer lymphedema dynamic prediction system, which comprises the following parts: a data collection module for dynamically collecting social demographic characteristics, physiological data, disease and treatment information, life habit data and psychosocial data of breast cancer patients; a data preprocessing module electrically connected with the data collection module, which is used for cleaning and processing original data, including abnormal value inspection and processing, data merging and variable conversion and missing value interpolation, so as to ensure the accuracy and integrity of the data; a variable screening module electrically connected with the data preprocessing module, which screens important static and dynamic variables based on evidence-based evidence, expert opinions and statistical analysis methods, and constructs a dynamic prediction model; a dynamic joint prediction model module electrically connected with the variable screening module, which adopts a joint survival analysis model to dynamically predict the lymphedema occurrence risk of the patient through a longitudinal submodel and a survival submodel.

2. The breast cancer related lymphedema dynamic risk prediction system of claim 1, wherein, The data collection module specifically comprises the following parts: a social demographic data collection unit for collecting social demographic information of the patient, including age, educational level, marital status, economic status and medical insurance type; a physiological data collection unit for collecting real-time lymphedema-related physiological data of the patient, including body mass index and bilateral upper limb circumference; a clinical data collection unit for collecting real-time clinical treatment information of the patient, including tumor maximum diameter, breast surgery type, axillary surgery type, number of cleaned lymph nodes, number of positive lymph nodes, postoperative complication condition, received chemotherapy scheme and postoperative radiotherapy information; a life habit data collection unit for monitoring self-management behavior and life habit of the patient related to lymphedema, including lymphedema self-management behavior, lymphedema risk and protection behavior and exercise behavior; a psychosocial data collection unit for collecting psychosocial variables related to lymphedema of the patient, including disease perception, anxiety, depression, self-efficacy and medical social support. 3.The breast cancer related lymphedema dynamic risk prediction system of claim 2, wherein, The data preprocessing module specifically comprises the following parts: a repeated data inspection and deduplication unit for checking whether there are repeated records in the data set, deleting the repeated data records and ensuring that each data point is unique; an abnormal value detection and processing unit for identifying abnormal values in the data by using Z-score detection and replacing the abnormal values with the mean value of the variable; a data normalization and standardization unit for normalizing the data, converting different scale features into the same range and converting the data into a standard normal distribution; a missing value processing unit for filling the missing values based on mean value interpolation, and if the missing value proportion is greater than 30%, the data point is combined with the adjacent data point and deleted; a data merging and connecting unit for merging data from different data sources and different collection time points, and ensuring the integrity of the data.

4. The breast cancer related lymphedema dynamic risk prediction system of claim 3, wherein, The variable screening module specifically comprises the following parts: a feature selection algorithm unit for evaluating each variable separately by using Cox single factor analysis and generalized estimation equation, and selecting variables that have a significant influence on the target variable; Static and dynamic variable identification unit: select features that do not change over time as static features, including user's gender, age, education level and region, select features that change over time as time series data, including user's body mass index, self-management behavior and lifestyle, mental status, exercise behavior, etc. Variable importance verification and cross-validation unit: use cross-validation techniques to evaluate the performance of selected features in the prediction model, verify the effectiveness of feature selection, evaluate the contribution of different features to the prediction results by building a preliminary model, and further filter and verify the optimal feature set.

5. The breast cancer related lymphedema dynamic risk prediction system of claim 4, wherein, The use of Pearson correlation coefficient to evaluate the correlation between variables, remove the variables with low correlation with the target variable Specifically includes: Wherein, the Pearson correlation coefficient formula is: where r is the Pearson correlation coefficient, x i y i is the data point for the two variables, is the mean of the variables x and y.

6. The breast cancer related lymphedema dynamic risk prediction system of claim 5, wherein, Each variable is evaluated by statistical significance test, and the variable with significant influence on the target variable is selected Specifically includes: Wherein, the statistical significance test formula is: wherein is the mean of two samples, is the variance of two samples, n1n2is the sample size of two samples, t is the significance, if the t value is less than 0.05, the feature has a significant impact on the target variable, the feature is retained.

7. The breast cancer related lymphedema dynamic risk prediction system of claim 6, wherein, The dynamic joint prediction model module specifically includes: Longitudinal sub-model construction unit: based on the time series data of patients during follow-up, longitudinal sub-model is established, the model captures the dynamic changes of patient's lymphedema related conditions at different time points, including changes of patient's body mass index, disease perception, anxiety and depression, self-efficacy, medical social support and self-management behavior, longitudinal sub-model is trained, and dynamic change factors affecting lymphedema occurrence are identified by analyzing patient follow-up data; Survival sub-model construction unit: Cox proportional hazards model is used to analyze the risk of patients developing lymphedema, and the risk of each patient developing lymphedema in the future is obtained based on the output of Cox proportional hazards model.

8. The breast cancer related lymphedema dynamic risk prediction system of claim 7, wherein, The dynamic joint prediction model module further includes: Joint model construction unit: combining longitudinal sub-model and survival sub-model, using joint survival analysis method, integrating information from two sub-models, considering the mutual relationship of longitudinal data and survival data; Model evaluation and optimization unit: using C-index consistency index, Brier calibration score, sensitivity, specificity, positive predictive value, negative predictive value, balanced accuracy and clinical decision curve to evaluate the prediction performance and clinical practicability of joint survival analysis model; Dynamic prediction and risk assessment unit: based on the trained joint survival analysis model, dynamic prediction is carried out, patient's lymphedema risk is updated in real time, patient's individual risk assessment is carried out, and patient's survival curve is generated; Wherein, the dynamic prediction formula is: wherein is the instantaneous risk of the event occurring at time t given the covariate X t h0(t) is the baseline hazard function, exp is the exponential function, and β1, β2, β p are regression coefficients, X t1 , X t2 , X tp is the value of the specific covariate at time t; Expert feedback and adjustment unit: according to the feedback of doctors and experts in the field, the model is further adjusted, the structure, input features and training process of the model are continuously optimized combined with the feedback of experts and the actual performance of the model; Visualization and report generation unit: generate visual report of patient's lymphedema risk, including survival curve and risk probability curve chart.

9. The breast cancer related lymphedema dynamic risk prediction system of claim 8, wherein, The longitudinal sub-model construction unit specifically includes: where y i (t) denotes the longitudinal outcome observed for the i-th subject at time point t, ε i (t) is the measurement error, m i (t) denotes the true, unobserved longitudinal outcome, affected by measurement error, β, b i are fixed and random effects coefficients, respectively, are row vectors of the fixed and random effects matrices, respectively, b i ~ N(0, D) denotes the assumption that b i follows a multivariate normal distribution with mean 0 and variance D, (t) ~ N(0, σ 2 ) denotes the assumption that ε i (t) follows a multivariate normal distribution with mean 0 and variance σ 2 .

10. The breast cancer related lymphedema dynamic risk prediction system of claim 9, wherein, The survival sub-model construction unit specifically includes: Wherein, the Cox proportional hazards model formula is: In the formula, h i (t|M i (t),ω i Let represent the risk function of individual i at time t. Let be the instantaneous risk function, describing the instantaneous probability that an event will occur in individual i near time t under given conditions. Let dt represent the instantaneous risk rate as the time interval shrinks infinitely. Let dt represent the length of the time interval. The limit of instantaneous probability is defined as dt→0. This represents the event time of individual i when the following conditions are met. The conditional probability that the value falls within the interval (t+dt). Indicates the time t, h of an individual's survival. i (t|M i (t),ω i )=h0(t)exp{γ T ω i +αm i (t)},t>0 is an extension of the Cox proportional hazards model in the joint model, used to analyze the association between longitudinal data and survival data, and to quantify the longitudinal process m. i (t) How the risk of the outcome event is affected.

Images