A chronic disease trend risk fusion analysis method and system based on timing feature optimization

Abstract

The present application relates to a kind of based on timing feature optimization's chronic disease trend risk fusion analysis method, obtains patient diagnosis and treatment data, it is filtered using course of disease, three-dimensional quantification standard of completeness, interfering disease, exports structured original data set.Construct corresponding individualized course of disease time axis of different time nodes, exports standardization individual data set.Standardization individual data set is analyzed, sampling verification strengthens segmented boundary, exports individualized stage division result.Each stage index data is modeled, extracts timing trend feature set.Integrate timing trend feature set and clinical confounding factors, construct individualized trend and risk association model.Statistics and clinical verification are used to verify individualized trend and risk association model, and export qualified model is verified.Based on qualified model, set risk threshold in stage, using visual tool and clinical intervention are connected, and output chronic disease management report.

Description

Technical Field

[0001] This invention relates to the field of chronic disease management and medical data analysis technology, and in particular to a trend risk fusion analysis method and system applicable to chronic diseases. Background Technology

[0002] Chronic diseases are characterized by long disease cycles, gradual disease progression, significant individual heterogeneity, and large differences in pathological characteristics at different disease stages. The core requirement for their diagnosis and management is to accurately capture the transition points of disease stages, the dynamic trends of clinical indicators, and the patterns of disease risk progression, thereby providing a scientific and reliable basis for developing individualized intervention plans. Currently, in the clinical analysis and management of chronic diseases such as type 2 diabetes, hypertension, and chronic kidney disease, the mainstream methods still rely on traditional data analysis algorithms to achieve disease course segmentation, trend modeling, and risk assessment. However, existing methods and core algorithms have many limitations in practical clinical applications and cannot meet the needs of precise and individualized diagnosis and management of chronic diseases. Specific problems are reflected in the following four aspects: Traditional data imputation algorithms use a uniform interpolation method to handle missing indicators, failing to differentiate between data sparsity and trend strength at different stages of chronic diseases, such as subclinical, clinical, and complication phases. Furthermore, they ignore the intrinsic correlations between various clinical indicators, resulting in low imputation accuracy and significant deviations between the imputation results and clinical reality, severely impacting the reliability of subsequent modeling and analysis. Simultaneously, the lack of clear quantitative standards for screening clinical diagnosis and treatment data easily leads to the inclusion of low-quality, highly interfering case data, resulting in insufficient relevance and effectiveness of the analyzed samples.

[0003] Traditional PELT-based segmentation algorithms rely solely on the statistical characteristics of data for disease segmentation, deviating from the clearly defined clinical thresholds and clinical event anchors in chronic disease treatment guidelines. The segmentation results lack clinical rationality, and the algorithm uses a fixed penalty coefficient for calculation, which cannot adapt to the individual heterogeneous data of different patients. This can easily lead to over-segmentation or under-segmentation, resulting in poor stability of the disease stage boundary points and difficulty in accurately reflecting the actual disease progression status of patients.

[0004] Traditional staged regression modeling simply distinguishes between different model types without incorporating individual patient intervention response differences. Furthermore, the model's loss function is fixed and not bound to clinical critical thresholds, leading to a disconnect between the extracted clinical indicator trends and actual clinical scenarios. Simultaneously, the model fails to differentiate its adaptation to the changing patterns of indicators at different stages of chronic diseases, exhibiting insufficient ability to capture different trends such as slow changes, complex fluctuations, and accelerated evolution, resulting in model fitting accuracy that falls short of clinical analysis needs.

[0005] Traditional Markov multistate models rely solely on patients' baseline characteristics to calculate fixed state transition probabilities, failing to integrate dynamic trends of clinical indicators and the effects of clinical interventions. This makes them unable to reflect the dynamic changes in disease progression and the impact of interventions on risk. Furthermore, the models do not adequately correct for confounding clinical factors such as age, gender, and comorbidities, resulting in a single dimension of risk assessment, significant biases in the assessment results, and difficulty in accurately achieving individualized risk warnings for chronic diseases.

[0006] In summary, existing chronic disease analysis methods have not yet formed a complete algorithm system adapted to the characteristics of clinical data and the progressive, phased evolution of chronic diseases. They suffer from core problems such as a disconnect between algorithms and clinical diagnosis and treatment scenarios, poor individual adaptability, insufficient robustness of analysis results, and difficulties in clinical translation and application. They are unable to achieve closed-loop management of the entire process from data processing to model building and then to clinical intervention. Summary of the Invention

[0007] The present invention aims to provide a method for fusion analysis of chronic disease trend risk based on time-series feature optimization, so as to overcome the shortcomings of the existing technology. The technical problem to be solved by the present invention is achieved through the following technical solution.

[0008] A method for chronic disease trend risk fusion analysis based on time-series feature optimization includes the following steps: S1: Obtain patient diagnosis and treatment data, and filter it using three quantitative criteria: disease course, completeness, and interfering diseases, and output the structured raw dataset D; S2: A phased adaptive data imputation algorithm is used to process outliers and missing values ​​in the structured original dataset D, constructing individualized disease progression timelines corresponding to different time points, and outputting standardized individual datasets. ; S3: Using the Clinical Constraints and Individual Calibration PELT algorithm on standardized individual datasets. The analysis was performed, and the segmentation boundaries were reinforced using Bootstrap sampling validation, resulting in individualized stage segmentation results. P ; S4: Based on the phase division results P A stage-and-individual dual-fit regression algorithm is used to model the indicator data at each stage and extract the time-series trend feature set. F and intervention intensity factor ; S5: Employs a trend-and-state fusion Markov multi-state model, integrating time-series trend feature sets. F Clinical confounding factors and intervention intensity factors Construct an individualized trend-risk correlation model M; S6: Validate the individualized trend-risk association model M using both statistical indicators and clinical expert scores, and output a validated model. ; S7: Based on the validation qualification model Set phased risk thresholds, use visualization tools and link them with clinical interventions to output chronic disease management reports that include disease progression trends, risk levels and individualized intervention recommendations.

[0009] Preferably, the quantitative standards for the three dimensions of disease course, completeness, and interfering diseases are as follows: the disease course is not less than three years, which is the time from the first abnormality of the patient's indicators to the time of data collection; the data completeness is not less than 70%, which is the ratio of the number of valid indicator records to the total number of indicators that should be recorded, and patients with serious interfering diseases are excluded; the structured original dataset D contains information on four aspects: basic patient information, multi-time point test indicators, diagnosis and complication records, and intervention measures.

[0010] Preferably, patients are divided into cohorts based on the same chronic disease type, disease stage, and age group. The mean values ​​of indicators for each cohort are calculated using multi-time-point test indicators from the structured raw dataset D. m and standard deviation s A phased adaptive data imputation algorithm was adopted to process outliers and missing values ​​in the structured original dataset D in stages: subclinical, clinical, and complication phases. Different imputation methods were used in each stage. Specifically, this included: according to Identify outliers, among which For a single indicator detection value, These are the mean values ​​of indicators for patients in the same cohort. The standard deviation of indicators for patients in the same cohort; Once identified outliers are clinically verified to be non-clinically specific, they are replaced with the mean of the same cohort index; if they are clinically specific, the original data are retained and labeled. The compensation will be carried out in stages, specifically including: For patients in the subclinical phase, values ​​should be supplemented using the following methods:

[0011] in, The results are for the missing indicators in subclinical patients. For the first The weights of each related indicator, For the number of related indicators, For the first Number of valid samples for each indicator For the first Among the related indicators, the first one is... Valid detection values ​​for each sample; For patients in the clinical phase, values ​​are supplemented using the following methods:

[0012] in, The results are for the completion of missing indicators for patients in the clinical phase. , Missing time points The index values ​​at two consecutive valid time points. For the time points corresponding to the missing values, , Missing time points Adjacent valid time points As the target indicator, As a clinical reference indicator, The correlation coefficient between the target indicator and the reference indicator ranges from 0.7 to 1.0. For patients in the complication phase, supplementation should be performed in the following ways:

[0013]

[0014] in, The results are for supplementing missing indicators in patients with complications. , Missing time points The index values ​​at two consecutive valid time points. For the time points corresponding to the missing values, 、 Missing time points Adjacent valid time points This is a trend strength adjustment factor, fixed at a value of 0.3. The slope of the trend of the complication period index is calculated by the difference between the index values ​​and the time difference between adjacent effective time points. It is used to quantify the degree of accelerated change and constrain the trend of the complementary value.

[0015] Preferably, the construction of individualized disease progression timelines corresponding to different time points outputs standardized individual datasets. include: Starting with the patient's first abnormal indicator as T0, and linking the diagnosis node T1, the complication occurrence node T2, and the intervention and adjustment node T3, an individualized disease progression timeline corresponding to different time points is constructed, forming a visualized individual disease progression trajectory. After standardization and unit unification, a standardized individual dataset is formed. .

[0016] Preferably, the clinically constrained and individually calibrated PELT algorithm is used to process the standardized individual dataset. The analysis was performed, and the segmentation boundaries were reinforced using Bootstrap sampling validation, resulting in individualized stage segmentation results. P include: The total cost function is calculated as follows:

[0017] in, Minimize the cost of the subsequence from segments 1 to n of the full sequence. For sequence 1 to The cumulative cost of the segment For stage-adaptive loss function, For individualized penalty coefficients, This refers to the individual indicator volatility coefficient. These are the stage weighting coefficients. This is an indicator function; it takes the value 1 if the subsequence contains clinical feature anchors, and 0.8 otherwise. This is a set of clinical feature anchor points, comprising key clinical events reflecting changes in the course of chronic diseases, including patient diagnosis milestones, complication occurrence milestones, and intervention adjustment milestones. The preset number of disease segmentation is 3 segments. The final number of segments is determined through cost function optimization. The total number of data points in the entire sequence, taken from... The length of the time series; The segment validity is verified using the following methods:

[0018] in, For the first The validity flag for each segment boundary is used to filter boundaries that do not conform to clinical logic; 1 indicates validity, and 0 indicates invalidity. For the predicted first The standard error of the estimated boundary point for each stage was calculated through 1000 repeated samplings with replacement, with the sample size remaining consistent with the original sample size in each sampling. This is a clinical reference threshold. This is the threshold tolerance, with a value ranging from 5% to 10%. Boundary point Changes in indicators before and after The clinically significant threshold is the minimum change in an indicator that is determined according to the guidelines for the diagnosis and treatment of chronic diseases and can reflect changes in the clinical significance of the indicator. The individual indicator volatility coefficient is obtained through the following method. :

[0019] in, For individual indicator standard deviation, This represents the mean of individual indicators; The confidence interval was calculated using Bootstrap sampling with replacement for 1000 iterations, with the sample size remaining the same as the original sample size for each iteration. The confidence interval is ( ),in The standard error of the boundary point estimate is used to adjust the boundary to within the confidence interval, and the individualized stage segmentation result is output. and the time boundaries of each stage, among which P 1 is the subclinical phase. P 2 represents the clinical phase. P 3 is the complication period.

[0020] Preferably, the stage-and-individual dual-fit regression algorithm is used to model the indicator data of each stage and extract the time-series trend feature set. F include: for P For stage 1 patients, the predicted values ​​of their target indicators for this stage are calculated using the following method:

[0021] in, For the first The patient P The predicted values ​​of the target indicators for Phase 1 No. Patient P The trend slope in stage 1, For the first Patient P The regression intercept in stage 1, For time variables, For random error term, for P Clinical threshold for Phase 1; for P For stage 2 patients, the predicted values ​​of their target indicators for this stage are calculated using the following method:

[0022] in, For the first Patient P The predicted values ​​of the target indicators for Phase 2, For the first Patient P The coefficients of the quadratic term in the second stage, For the first Patient P The coefficient of the first-order term in the second stage, For the first Patient P The two-stage regression intercept, For time variables, This is the random error term; for P For stage 3 patients, the predicted values ​​of their target indicators for that stage are calculated using the following method:

[0023]

[0024] in, For the first Patient P The predicted values ​​of the target indicators for the three stages. 、 、 The first Patient P The coefficients of the quadratic term, the coefficients of the linear term, and the intercept in the three stages. For time variables, The intervention impact coefficient, For the first The intervention intensity factor for each patient ranges from 0 to 1 and is quantified according to medication dosage / frequency. For random error term, For weighted loss function, For the total number of patients, This is the outlier weighting coefficient. , The actual value and the model prediction value of the indicator; By integrating model parameters and volatility coefficients from each stage, a time-series trend feature set is formed. :

[0025] in, They are respectively P The slope, intercept, and fluctuation coefficient of stage 1. for P The quadratic coefficient, linear coefficient, intercept, and fluctuation coefficient of the two-stage term. for P The coefficients of the quadratic term, the coefficients of the linear term, the intercept, and the fluctuation coefficient in the three stages.

[0026] Preferably, the method employs a trend-state fusion Markov multi-state model to integrate time-series trend feature sets. F In conjunction with clinical confounding factors, an individualized trend-risk association model M is constructed, including: The dynamic state transition probability is calculated in the following way:

[0027]

[0028] in, For time Internal state Transfer to The dynamic probability, For follow-up time, For disease course state variables, For individual intervention intensity factors, From arrive Baseline transition probability, Let be the trend feature influence function, where the input parameters are... From time series trend feature set , The intervention impact coefficient reuses the value from step S4; The trend-weighted risk ratio is calculated as follows:

[0029] in, This represents the individualized risk ratio corresponding to the trend characteristics. For the risk function of the high-trend risk group, For the risk function of the low-trend risk group, based on the time series trend feature set Based on the core features, the high-risk group is defined as a² ≥ 0.05 and CV² ≥ 0.09, while the rest are low-risk groups. The baseline risk ratio, For trend feature weights, For the value of a single trend feature, This represents the mean of this trend characteristic among patients in the same stage. The standard deviation of this trend characteristic for patients in the same stage. A collection of mixed factors; The confounding factor correction formula is calculated as follows:

[0030] in, This is the corrected individualized risk function. As the benchmark risk function, For a collection of mixed factors, This represents the regression coefficient vector of trend characteristics. For time series trend feature set, This represents the vector of regression coefficients for confounding factors. , All data were obtained by fitting a Cox proportional hazards regression model. The fitted data consisted of phased indicator data and clinical confounding factor data from a standardized individual dataset. The output includes an individualized trend and risk association model M, which includes dynamic state transition probability, trend-weighted hazard ratio, and adjusted individualized risk. The individualized trend and risk association model is a risk assessment model that integrates time-series trend features, clinical confounding factors, and intervention intensity factors, and can output dynamic state transition probability and trend-weighted hazard ratio.

[0031] Preferably, the individualized trend-risk association model M is validated using statistical and clinical validation, and a validated model is output. include: First, stratified balanced sampling is used to divide the samples into training and test sets in a 7:3 ratio. Then, the effectiveness of the model is verified by a fusion validation metric CS. The fusion validation metric CS is as follows:

[0032] in, To improve model discriminative power, based on The predicted results are calculated in relation to the actual stage transition. Clinical applicability score; Clinical applicability score Calculated in the following way:

[0033] in, For the number of clinical experts, For risk threshold clinical effectiveness scoring, The reasonableness score for the interpretation of the results is rated, with a value range of 0 to 1. The fusion validation index CS and clinical applicability score The model's discrimination index (AUC) and calibration score are used for validation. Once the validation is successful, the validated model is output. The validation criteria included the fusion validation index CS and the clinical applicability score. The model discrimination AUC is not less than 0.8 and the calibration is greater than 0.05.

[0034] Preferably, the visualization tool includes: Based on time-series trend feature set F The indicator trend tracking module for drawing time series trend lines is based on a validated model. The module outputs a risk prediction curve and confidence interval for the occurrence of complications, as well as an intervention recommendation module that automatically matches quantitative intervention plans based on risk thresholds.

[0035] This invention also provides a chronic disease trend risk fusion analysis system based on time-series feature optimization, comprising: The multi-source data acquisition module acquires patient diagnosis and treatment data, and filters it using three quantitative standards: disease course, completeness, and interfering diseases, and outputs a structured raw dataset D. The individual data standardization and timeline construction module employs a phased adaptive data imputation algorithm to process outliers and missing values ​​in the structured original dataset D, constructing individualized disease progression timelines corresponding to different time points, and outputting standardized individual datasets. ; The disease stage identification module uses the Clinical Constraints and Individual Calibration (PELT) algorithm on a standardized individual dataset. The analysis was performed, and the segmentation boundaries were reinforced using Bootstrap sampling validation, resulting in individualized stage segmentation results. P ; The feature extraction module, based on the stage division results P A stage-and-individual dual-fit regression algorithm is used to model the indicator data at each stage and extract the time-series trend feature set. F and intervention intensity factor ; The analysis and fusion module employs a trend and state fusion Markov multi-state model to integrate time series trend feature sets. F Clinical confounding factors and intervention intensity factors Construct an individualized trend-risk correlation model M; The validation module uses both statistical indicators and clinical expert scores to validate the individualized trend-risk association model M, outputting a validated model. ; Output module, based on the validation success model Set phased risk thresholds, use visualization tools and link them with clinical interventions to output chronic disease management reports that include disease progression trends, risk levels and individualized intervention recommendations.

[0036] Compared with existing technologies, the chronic disease trend risk fusion analysis method based on time-series feature optimization of the present invention has the following beneficial effects: 1. Each core component incorporates a proprietary optimization algorithm, and all algorithms are aligned with the phased evolution of chronic diseases and the characteristics of clinical data, overcoming the limitations of traditional one-size-fits-all algorithms. The phased adaptive data compensation algorithm solves the problem of low accuracy in uniform compensation; the clinical constraint and individual calibration PELT algorithm achieves precise clinical and individualized segmentation of disease stages; the phase and individual dual-fit regression algorithm improves the fitting accuracy of trend feature extraction; and the trend and state fusion Markov multi-state model solves the problems of single dimensions and large biases in traditional risk assessment. The accuracy of data processing, modeling, and assessment in each stage is significantly improved, fundamentally solving the core problems of traditional algorithms being disconnected from clinical practice and having poor adaptability.

[0037] 2. Using a core technology path of data input, processing, analysis, fusion, validation, and transformation, the system comprises seven logically progressive and interconnected steps: multi-source longitudinal data acquisition, individualized data standardization, disease stage identification, time-series trend modeling, trend and risk fusion modeling, model dual-effect validation, risk warning, and clinical application. Parameters are shared and results are interchangeable among the various optimization algorithms, forming an integrated precision analysis technology system for chronic disease progression. This system achieves a closed-loop process from raw clinical data to standardized processing, from model construction to dual-effect validation, and from risk assessment to the transformation into clinical intervention tools. It avoids the fragmentation and disconnect between stages of traditional technologies, ensuring the continuity and effectiveness of technology implementation.

[0038] 3. The algorithm optimization process incorporates a wealth of prior clinical knowledge, clearly defined clinical thresholds and intervention logic from chronic disease treatment guidelines. The model output exhibits both statistical robustness and clinical rationality, significantly improving interpretability and operability. The visualization tool developed based on the validated model seamlessly integrates with hospital HIS / LIS systems, updating data at the same frequency as patient follow-up periods. It automatically matches quantitative intervention plans based on phased risk thresholds, achieving a closed-loop clinical management system encompassing risk identification, intervention implementation, and effect tracking. This overcomes the limitations of traditional chronic disease early warning systems that emphasize identification over intervention. Furthermore, the generated precise chronic disease management report can be directly embedded into existing hospital chronic disease management systems, providing clinicians with intuitive and specific diagnostic and treatment references without complex secondary processing. This directly serves clinical practice, improving the level of individualized chronic disease management and possessing broad clinical application value.

[0039] 4. This invention achieves individualized adaptation at every stage, from data screening and phase division to risk assessment and intervention recommendations. It accurately captures differences in disease progression, indicator trends, and intervention responses among different patients, breaking away from the traditional group-based and standardized model of chronic disease management and truly realizing individualized and precise management of chronic diseases. Simultaneously, the visualization tools and automated intervention plan matching significantly reduce the workload of clinicians in manual data analysis, improve the efficiency of chronic disease diagnosis and management, help alleviate the workload of clinical chronic disease management, and promote the development of chronic disease management towards intelligence and precision. Detailed Implementation

[0040] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the embodiments.

[0041] A method for chronic disease trend risk fusion analysis based on time-series feature optimization includes the following steps: S1: Obtain patient diagnosis and treatment data, and filter it using three quantitative standards: disease course, completeness, and interfering diseases, and output a structured raw dataset D.

[0042] The quantitative standards for the three dimensions of disease course, completeness, and interfering diseases are as follows: the disease course is no less than three years, which is the time from the first abnormality of the patient's indicators to the time of data collection; the data completeness is no less than 70%, which is the ratio of the number of valid indicator records to the total number of indicators that should be recorded, and patients with serious interfering diseases (including acute or malignant diseases that affect the assessment of chronic disease course, such as clearly diagnosed malignant tumors, acute myocardial infarction, and severe infections) are excluded; the structured original dataset D contains information on four aspects: patient basic information, multi-time point test indicators, diagnosis and complication records, and intervention measures.

[0043] S2: A phased adaptive data imputation algorithm is used to process outliers and missing values ​​in the structured original dataset D, constructing individualized disease progression timelines corresponding to different time points, and outputting standardized individual datasets. .

[0044] Patients were divided into cohorts based on the same chronic disease type, disease stage, and age group. The mean values ​​of indicators within each cohort were calculated using multi-timepoint test indicators from the structured raw dataset D. m and standard deviation s A phased adaptive data imputation algorithm was adopted to process outliers and missing values ​​in the structured original dataset D in stages: subclinical, clinical, and complication phases. Different imputation methods were used in each stage. Specifically, this included: according to Identify outliers, among which For a single indicator detection value, These are the mean values ​​of indicators for patients in the same cohort. The standard deviation of indicators for patients in the same cohort is given.

[0045] Once identified outliers are clinically verified to be non-clinically specific, they are replaced with the mean of the same cohort index; if they are clinically specific, the original data are retained and labeled. The compensation will be carried out in stages, specifically including: For patients in the subclinical phase, values ​​should be supplemented using the following methods:

[0046] in, The results are for the missing indicators in subclinical patients. For the first The weights of each related indicator, For the number of related indicators, For the first Number of valid samples for each indicator For the first Among the related indicators, the first one is... Valid detection values ​​for each sample; For patients in the clinical phase, values ​​are supplemented using the following methods:

[0047] in, The results are for the completion of missing indicators for patients in the clinical phase. , Missing time points The index values ​​at two consecutive valid time points. For the time points corresponding to the missing values, , Missing time points Adjacent valid time points As the target indicator, As a clinical reference indicator, The correlation coefficient between the target indicator and the reference indicator ranges from 0.7 to 1.0. For patients in the complication phase, supplementation should be performed in the following ways:

[0048]

[0049] in, The results are for supplementing missing indicators in patients with complications. , Missing time points The index values ​​at two consecutive valid time points. For the time points corresponding to the missing values, 、 Missing time points Adjacent valid time points This is a trend strength adjustment factor, fixed at a value of 0.3. The slope of the trend of the complication period index is calculated by the difference between the index values ​​and the time difference between adjacent effective time points. It is used to quantify the degree of accelerated change and constrain the trend of the complementary value.

[0050] Starting with the patient's first abnormal indicator as T0, and linking the diagnosis node T1, the complication occurrence node T2, and the intervention and adjustment node T3, an individualized disease progression timeline corresponding to different time points is constructed, forming a visualized individual disease progression trajectory. After standardization and unit unification, a standardized individual dataset is formed. .

[0051] S3: Using the Clinical Constraints and Individual Calibration PELT algorithm on standardized individual datasets. The analysis was performed, and the segmentation boundaries were reinforced using Bootstrap sampling validation, resulting in individualized stage segmentation results. P .

[0052] The total cost function is calculated as follows:

[0053] in, Minimize the cost of the subsequence from segments 1 to n of the full sequence. For sequence 1 to The cumulative cost of the segment For stage-adaptive loss function, For individualized penalty coefficients, This refers to the individual indicator volatility coefficient. These are the stage weighting coefficients. This is an indicator function; it takes the value 1 if the subsequence contains clinical feature anchors, and 0.8 otherwise. This is a set of clinical feature anchor points, comprising key clinical events reflecting changes in the course of chronic diseases, including patient diagnosis milestones, complication occurrence milestones, and intervention adjustment milestones. The preset number of disease segmentation is 3 segments. The final number of segments is determined through cost function optimization. The total number of data points in the entire sequence, taken from... The length of the time series; The segment validity is verified using the following methods:

[0054] in, For the first The validity flag for each segment boundary is used to filter boundaries that do not conform to clinical logic; 1 indicates validity, and 0 indicates invalidity. For the predicted first The standard error of the estimated boundary point for each stage was calculated through 1000 repeated samplings with replacement, with the sample size remaining consistent with the original sample size in each sampling. This is a clinical reference threshold. This is the threshold tolerance, with a value ranging from 5% to 10%. Boundary point Changes in indicators before and after The clinically significant threshold is the minimum change in an indicator that is determined according to the guidelines for the diagnosis and treatment of chronic diseases and can reflect changes in the clinical significance of the indicator. The individual indicator volatility coefficient is obtained through the following method. :

[0055] in, For individual indicator standard deviation, This represents the mean of individual indicators; The confidence interval was calculated using Bootstrap sampling with replacement for 1000 iterations, with the sample size remaining the same as the original sample size for each iteration. The confidence interval is ( ),in The standard error of the boundary point estimate is used to adjust the boundary to within the confidence interval, and the individualized stage segmentation result is output. and the time boundaries of each stage, among which P 1 is the subclinical phase. P 2 represents the clinical phase. P 3 is the complication period.

[0056] S4: Based on the phase division results P A stage-and-individual dual-fit regression algorithm is used to model the indicator data at each stage and extract the time-series trend feature set. F and intervention intensity factor .

[0057] for P For stage 1 patients, the predicted values ​​of their target indicators for this stage are calculated using the following method:

[0058] in, For the first The patient P The predicted values ​​of the target indicators for Phase 1 No. Patient P The trend slope in stage 1, For the first Patient P The regression intercept in stage 1, For time variables, For random error term, for P Clinical threshold for Phase 1; for P For stage 2 patients, the predicted values ​​of their target indicators for this stage are calculated using the following method:

[0059] in, For the first Patient P The predicted values ​​of the target indicators for Phase 2, For the first Patient P The coefficients of the quadratic term in the second stage, For the first Patient PThe coefficient of the first-order term in the second stage, For the first Patient P The two-stage regression intercept, For time variables, This is the random error term; for P For stage 3 patients, the predicted values ​​of their target indicators for that stage are calculated using the following method:

[0060]

[0061] in, For the first Patient P The predicted values ​​of the target indicators for the three stages. 、 、 The first Patient P The coefficients of the quadratic term, the coefficients of the linear term, and the intercept in the three stages. For time variables, The intervention impact coefficient, For the first The intervention intensity factor for each patient ranges from 0 to 1 and is quantified according to medication dosage / frequency. For random error term, For weighted loss function, For the total number of patients, This is the outlier weighting coefficient. , The actual value and the model prediction value of the indicator; By integrating model parameters and volatility coefficients from each stage, a time-series trend feature set is formed. :

[0062] in, They are respectively P The slope, intercept, and fluctuation coefficient of stage 1. for P The quadratic coefficient, linear coefficient, intercept, and fluctuation coefficient of the two-stage term. for P The coefficients of the quadratic term, the coefficients of the linear term, the intercept, and the fluctuation coefficient in the three stages.

[0063] S5: Employs a trend-and-state fusion Markov multi-state model, integrating time-series trend feature sets. F Clinical confounding factors and intervention intensity factors We will construct an individualized trend and risk correlation model M.

[0064] The dynamic state transition probability is calculated in the following way:

[0065]

[0066] in, For time Internal state Transfer to The dynamic probability, For follow-up time, For disease course state variables, For individual intervention intensity factors, From arrive Baseline transition probability, Let be the trend feature influence function, where the input parameters are... From time series trend feature set , The intervention impact coefficient reuses the value from step S4; The trend-weighted risk ratio is calculated as follows:

[0067] in, This represents the individualized risk ratio corresponding to the trend characteristics. For the risk function of the high-trend risk group, For the risk function of the low-trend risk group, based on the time series trend feature set Based on the core features, the high-risk group is defined as a² ≥ 0.05 and CV² ≥ 0.09, while the rest are low-risk groups. The baseline risk ratio, For trend feature weights, For the value of a single trend feature, This represents the mean of this trend characteristic among patients in the same stage. The standard deviation of this trend characteristic for patients in the same stage. A collection of mixed factors; The confounding factor correction formula is calculated as follows:

[0068] in, This is the corrected individualized risk function. As the benchmark risk function, For a collection of mixed factors, This represents the regression coefficient vector of trend characteristics. For time series trend feature set, This represents the vector of regression coefficients for confounding factors. , All data were obtained by fitting a Cox proportional hazards regression model. The fitted data consisted of phased indicator data and clinical confounding factor data from a standardized individual dataset. The output includes an individualized trend and risk association model M, which includes dynamic state transition probability, trend-weighted hazard ratio, and adjusted individualized risk. The individualized trend and risk association model is a risk assessment model that integrates time-series trend features, clinical confounding factors, and intervention intensity factors, and can output dynamic state transition probability and trend-weighted hazard ratio.

[0069] S6: Validate the individualized trend-risk association model M using both statistical indicators and clinical expert scores, and output a validated model. .

[0070] First, stratified balanced sampling is used to divide the samples into training and test sets in a 7:3 ratio. Then, the effectiveness of the model is verified by a fusion validation metric CS. The fusion validation metric CS is as follows:

[0071] in, To improve model discriminative power, based on The predicted results are calculated in relation to the actual stage transition. Clinical applicability score; Clinical applicability score Calculated in the following way:

[0072] in, For the number of clinical experts, For risk threshold clinical effectiveness scoring, The reasonableness score for the interpretation of the results is rated, with a value range of 0 to 1. The fusion validation index CS and clinical applicability score The model's discrimination index (AUC) and calibration score are used for validation. Once the validation is successful, the validated model is output. The validation criteria included the fusion validation index CS and the clinical applicability score. The model discrimination AUC is not less than 0.8 and the calibration is greater than 0.05.

[0073] S7: Based on the validation qualification model Set phased risk thresholds, use visualization tools and link them with clinical interventions to output chronic disease management reports that include disease progression trends, risk levels and individualized intervention recommendations.

[0074] The visualization tool includes: based on time-series trend feature sets. F The indicator trend tracking module for drawing time series trend lines is based on a validated model. The module outputs a risk prediction curve and confidence interval for the occurrence of complications, as well as an intervention recommendation module that automatically matches quantitative intervention plans based on risk thresholds.

[0075] This invention also provides a chronic disease trend risk fusion analysis system based on time-series feature optimization, comprising: The multi-source data acquisition module acquires patient diagnosis and treatment data, and filters it using three quantitative standards: disease course, completeness, and interfering diseases, and outputs a structured raw dataset D. The individual data standardization and timeline construction module employs a phased adaptive data imputation algorithm to process outliers and missing values ​​in the structured original dataset D, constructing individualized disease progression timelines corresponding to different time points, and outputting standardized individual datasets. ; The disease stage identification module uses the Clinical Constraints and Individual Calibration (PELT) algorithm on a standardized individual dataset. The analysis was performed, and the segmentation boundaries were reinforced using Bootstrap sampling validation, resulting in individualized stage segmentation results. P ; The feature extraction module, based on the stage division results P A stage-and-individual dual-fit regression algorithm is used to model the indicator data at each stage and extract the time-series trend feature set. F and intervention intensity factor ; The analysis and fusion module employs a trend and state fusion Markov multi-state model to integrate time series trend feature sets. F Clinical confounding factors and intervention intensity factors Construct an individualized trend-risk correlation model M; The validation module uses both statistical indicators and clinical expert scores to validate the individualized trend-risk association model M, outputting a validated model. ; Output module, based on the validation success model Set phased risk thresholds, use visualization tools and link them with clinical interventions to output chronic disease management reports that include disease progression trends, risk levels and individualized intervention recommendations.

[0076] It should be noted that the above detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0077] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments described in this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0078] Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or apparatus.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for chronic disease trend and risk fusion analysis based on time-series feature optimization, characterized in that, Includes the following steps: S1: Obtain patient diagnosis and treatment data, and filter it using three quantitative criteria: disease course, completeness, and interfering diseases, and output the structured raw dataset D; S2: A phased adaptive data imputation algorithm is used to process outliers and missing values ​​in the structured original dataset D, constructing individualized disease progression timelines corresponding to different time points, and outputting standardized individual datasets. ; S3: Using the Clinical Constraints and Individual Calibration PELT algorithm on standardized individual datasets. The analysis was performed, and the segmentation boundaries were reinforced using Bootstrap sampling validation, resulting in individualized stage segmentation results. P ; S4: Based on the phase division results P A stage-and-individual dual-fit regression algorithm is used to model the indicator data at each stage and extract the time-series trend feature set. F and intervention intensity factor ; S5: Employs a trend-and-state fusion Markov multi-state model, integrating time-series trend feature sets. F Clinical confounding factors and intervention intensity factors Construct an individualized trend-risk correlation model M; S6: Validate the individualized trend-risk association model M using both statistical indicators and clinical expert scores, and output a validated model. ; S7: Based on the validation qualification model Set phased risk thresholds, use visualization tools and link them with clinical interventions to output chronic disease management reports that include disease progression trends, risk levels and individualized intervention recommendations.

2. The chronic disease trend risk fusion analysis method based on time-series feature optimization according to claim 1, characterized in that: In step S1, the quantitative standards for the three dimensions of disease course, completeness, and interfering diseases are as follows: the disease course is not less than three years, which is the time from the first abnormality of the patient's indicators to the time of data collection; the data completeness is not less than 70%, which is the ratio of the number of valid indicator records to the total number of indicators that should be recorded, and patients with serious interfering diseases are excluded; the structured original dataset D contains information on four aspects: basic patient information, multi-time point test indicators, diagnosis and complication records, and intervention measures.

3. The chronic disease trend risk fusion analysis method based on time-series feature optimization according to claim 2, characterized in that: In step S2, patients are divided into cohorts based on the same chronic disease type, disease stage, and age group. The mean values ​​of the indicators for each cohort are calculated using the multi-time-point test indicators from the structured raw dataset D. μ and standard deviation σ A phased adaptive data imputation algorithm was adopted to process outliers and missing values ​​in the structured original dataset D in stages: subclinical, clinical, and complication phases. Different imputation methods were used in each stage. Specifically, this included: according to Identify outliers, among which For a single indicator detection value, These are the mean values ​​of indicators for patients in the same cohort. The standard deviation of indicators for patients in the same cohort; Once identified outliers are clinically verified to be non-clinically specific, they are replaced with the mean of the same cohort index; if they are clinically specific, the original data are retained and labeled. The compensation will be carried out in stages, specifically including: For patients in the subclinical phase, values ​​should be supplemented using the following methods: in, The results are for the missing indicators in subclinical patients. For the first The weights of each related indicator, For the number of related indicators, For the first Number of valid samples for each indicator For the first Among the related indicators, the first one is... Valid detection values ​​for each sample; For patients in the clinical phase, values ​​are supplemented using the following methods: in, The results are for the completion of missing indicators for patients in the clinical phase. , Missing time points The index values ​​at two consecutive valid time points. For the time points corresponding to the missing values, , Missing time points Adjacent valid time points As the target indicator, As a clinical reference indicator, The correlation coefficient between the target indicator and the reference indicator ranges from 0.7 to 1.

0. For patients in the complication phase, supplementation should be performed in the following ways: in, The results are for supplementing missing indicators in patients with complications. , Missing time points The index values ​​at two consecutive valid time points. For the time points corresponding to the missing values, 、 Missing time points Adjacent valid time points This is a trend strength adjustment factor, fixed at a value of 0.

3. The slope of the trend of the complication period index is calculated by the difference between the index values ​​and the time difference between adjacent effective time points. It is used to quantify the degree of accelerated change and constrain the trend of the complementary value.

4. The chronic disease trend risk fusion analysis method based on time-series feature optimization according to claim 3, characterized in that: In step S2, the individualized disease progression timelines corresponding to different time points are constructed, and standardized individual datasets are output. include: Starting with the patient's first abnormal indicator as T0, and linking the diagnosis node T1, the complication occurrence node T2, and the intervention adjustment node T3, an individualized disease progression timeline corresponding to different time points is constructed, forming a visualized individual disease progression trajectory. After standardization and unit unification, a standardized individual dataset is formed. .

5. The chronic disease trend risk fusion analysis method based on time-series feature optimization according to claim 1, characterized in that: In step S3, the clinically constrained and individually calibrated PELT algorithm is used to process the standardized individual dataset. The analysis was performed, and the segmentation boundaries were reinforced using Bootstrap sampling validation, resulting in individualized stage segmentation results. P include: The total cost function is calculated as follows: in, Minimize the cost of the subsequence from segments 1 to n of the full sequence. For sequence 1 to The cumulative cost of the segment For stage-adaptive loss function, For individualized penalty coefficients, This refers to the individual indicator volatility coefficient. These are the stage weighting coefficients. This is an indicator function; it takes the value 1 if the subsequence contains clinical feature anchors, and 0.8 otherwise. This is a set of clinical feature anchor points, comprising key clinical events reflecting changes in the course of chronic diseases, including patient diagnosis milestones, complication occurrence milestones, and intervention adjustment milestones. The preset number of disease segmentation is 3 segments. The final number of segments is determined through cost function optimization. The total number of data points in the entire sequence, taken from... The length of the time series; The segment validity is verified using the following methods: in, For the first The validity flag for each segment boundary is used to filter boundaries that do not conform to clinical logic; 1 indicates validity, and 0 indicates invalidity. For the predicted first The standard error of the estimated boundary point for each stage was calculated through 1000 repeated samplings with replacement, with the sample size remaining consistent with the original sample size in each sampling. This is a clinical reference threshold. This is the threshold tolerance, with a value ranging from 5% to 10%. Boundary point Changes in indicators before and after The clinically significant threshold is the minimum change in an indicator that is determined according to the guidelines for the diagnosis and treatment of chronic diseases and can reflect changes in the clinical significance of the indicator. The individual indicator volatility coefficient is obtained through the following method. : in, For individual indicator standard deviation, This represents the mean of individual indicators; The confidence interval was calculated using Bootstrap sampling with replacement for 1000 iterations, with the sample size remaining the same as the original sample size for each iteration. The confidence interval is ( ),in The standard error of the boundary point estimate is used to adjust the boundary to within the confidence interval, and the individualized stage segmentation result is output. and the time boundaries of each stage, among which P 1 is the subclinical phase. P 2 represents the clinical phase. P 3 is the complication period.

6. The chronic disease trend risk fusion analysis method based on time-series feature optimization according to claim 5, characterized in that: In step S4, the stage and individual dual-fit regression algorithm is used to model the indicator data of each stage and extract the time-series trend feature set. F include: for P For stage 1 patients, the predicted values ​​of their target indicators for this stage are calculated using the following method: in, For the first The patient P The predicted values ​​of the target indicators for Phase 1 No. Patient P The trend slope in stage 1, For the first Patient P The regression intercept in stage 1, For time variables, For random error term, for P Clinical threshold for Phase 1; for P For stage 2 patients, the predicted values ​​of their target indicators for this stage are calculated using the following method: in, For the first Patient P The predicted values ​​of the target indicators for Phase 2, For the first Patient P The coefficients of the quadratic term in the second stage, For the first Patient P The coefficient of the first-order term in the second stage, For the first Patient P The two-stage regression intercept, For time variables, This is the random error term; for P For stage 3 patients, the predicted values ​​of their target indicators for that stage are calculated using the following method: in, For the first Patient P The predicted values ​​of the target indicators for the three stages. 、 、 The first Patient P The coefficients of the quadratic term, the coefficients of the linear term, and the intercept in the three stages. For time variables, The intervention impact coefficient, For the first The intervention intensity factor for each patient ranges from 0 to 1 and is quantified according to medication dosage / frequency. For random error term, For weighted loss function, For the total number of patients, This is the outlier weighting coefficient. , The actual value and the model prediction value of the indicator; By integrating model parameters and volatility coefficients from each stage, a time-series trend feature set is formed. : in, They are respectively P The slope, intercept, and fluctuation coefficient of stage 1. for P The quadratic coefficient, linear coefficient, intercept, and fluctuation coefficient of the two-stage term. for P The coefficients of the quadratic term, the coefficients of the linear term, the intercept, and the fluctuation coefficient in the three stages.

7. The chronic disease trend risk fusion analysis method based on time-series feature optimization according to claim 6, characterized in that: In step S5, a trend-state fusion Markov multi-state model is adopted to integrate the time-series trend feature set. F In conjunction with clinical confounding factors, an individualized trend-risk association model M is constructed, including: The dynamic state transition probability is calculated in the following way: in, For time Internal state Transfer to The dynamic probability, For follow-up time, For disease course state variables, For individual intervention intensity factors, From arrive Baseline transition probability, Let be the trend feature influence function, where the input parameters are... From time series trend feature set , The intervention impact coefficient reuses the value from step S4; The trend-weighted risk ratio is calculated as follows: in, This represents the individualized risk ratio corresponding to the trend characteristics. For the risk function of the high-trend risk group, For the risk function of the low-trend risk group, based on the time series trend feature set Based on the core features, the high-risk group is defined as a² ≥ 0.05 and CV² ≥ 0.09, while the rest are low-risk groups. The baseline hazard ratio, For trend feature weights, For the value of a single trend feature, This represents the mean of this trend characteristic among patients in the same stage. The standard deviation of this trend characteristic for patients in the same stage. It is a collection of mixed factors; The confounding factor correction formula is calculated as follows: in, This is the corrected individualized risk function. As the benchmark risk function, For a collection of mixed factors, This represents the regression coefficient vector of trend characteristics. For time series trend feature set, This represents the vector of regression coefficients for confounding factors. , All data were obtained by fitting a Cox proportional hazards regression model. The fitted data consisted of phased indicator data and clinical confounding factor data from a standardized individual dataset. The output includes an individualized trend and risk association model M, which includes dynamic state transition probability, trend-weighted hazard ratio, and adjusted individualized risk. The individualized trend and risk association model is a risk assessment model that integrates time-series trend features, clinical confounding factors, and intervention intensity factors, and can output dynamic state transition probability and trend-weighted hazard ratio.

8. The chronic disease trend risk fusion analysis method based on time-series feature optimization according to claim 7, characterized in that: In step S6, the individualized trend-risk association model M is validated using statistical and clinical validation, and a validated model is output. include: First, stratified balanced sampling is used to divide the samples into training and test sets in a 7:3 ratio. Then, the effectiveness of the model is verified by a fusion validation metric CS. The fusion validation metric CS is as follows: in, To improve model discriminative power, based on The predicted results are calculated in relation to the actual stage transition. Clinical applicability score; Clinical applicability score Calculated in the following way: in, For the number of clinical experts, For risk threshold clinical effectiveness scoring, The reasonableness score for the interpretation of the results is rated, with a value range of 0 to 1. The fusion validation index CS and clinical applicability score The model's discrimination index (AUC) and calibration score are used for validation. Once the validation is successful, the validated model is output. The validation criteria included the fusion validation index CS and the clinical applicability score. The model discrimination AUC is not less than 0.8 and the calibration is greater than 0.

05.

9. The chronic disease trend risk fusion analysis method based on time-series feature optimization according to claim 8, characterized in that: In step S7, the visualization tool includes: Based on time-series trend feature set F The indicator trend tracking module for drawing time series trend lines is based on a validated model. The module outputs a risk prediction curve and confidence interval for the occurrence of complications, as well as an intervention recommendation module that automatically matches quantitative intervention plans based on risk thresholds.

10. A chronic disease trend risk fusion analysis system based on time-series feature optimization, characterized in that, include: The multi-source data acquisition module acquires patient diagnosis and treatment data, and filters it using three quantitative standards: disease course, completeness, and interfering diseases, and outputs a structured raw dataset D. The individual data standardization and timeline construction module employs a phased adaptive data imputation algorithm to process outliers and missing values ​​in the structured original dataset D, constructing individualized disease progression timelines corresponding to different time points, and outputting standardized individual datasets. ; The disease stage identification module uses the Clinical Constraints and Individual Calibration (PELT) algorithm on a standardized individual dataset. The analysis was performed, and the segmentation boundaries were reinforced using Bootstrap sampling validation, resulting in individualized stage segmentation results. P ; The feature extraction module uses a stage-and-individual dual-fit regression algorithm to model the indicator data at each stage and extract the time-series trend feature set. F ; The analysis and fusion module employs a trend and state fusion Markov multi-state model to integrate time series trend feature sets. F To incorporate clinical confounding factors, a personalized trend-risk association model M was constructed. The validation module uses both statistical indicators and clinical expert scores to validate the individualized trend-risk association model M, outputting a validated model. ; Output module, based on the validation success model Set phased risk thresholds, use visualization tools and link them with clinical interventions to output chronic disease management reports that include disease progression trends, risk levels and individualized intervention recommendations.

Images