Liver disease patient full-period management and risk layering service system

Abstract

The invention discloses a hepatopathy patient full-period management and risk layering service system, which belongs to the technical field of medical health data processing, and comprises a multi-source data acquisition and fusion module, a multi-dimensional risk layering and prediction module, an intelligent resource scheduling and path planning module and a closed-loop feedback and parameter optimization module, the system constructs a four-dimensional risk assessment system by integrating clinical data, laboratory indexes, iconography characteristics and remote monitoring data, generates a patient comprehensive risk level by applying an adaptive weighted fusion algorithm, dynamically schedules medical resources and formulates a personalized management scheme according to the risk level, and continuously optimizes model parameters through a closed-loop feedback mechanism, so that the risk assessment accuracy is improved. A deep coupling cooperative relationship is formed among the modules, the early recognition rate of high-risk patients is improved, the re-admission rate is reduced, the nursing efficiency is improved, the resource utilization rate is improved, and an effective technical solution is provided for complete-cycle accurate management of the liver disease patients.

Description

Technical Field

[0001] This invention belongs to the field of medical and health data processing technology, specifically involving a full-cycle management and risk stratification service system for liver disease patients, and particularly involving dynamic risk assessment, intelligent resource scheduling and closed-loop optimization technology for patients based on multi-dimensional clinical data. Background Technology

[0002] Chronic liver disease is a significant global public health issue, encompassing various types such as fatty liver disease associated with metabolic dysfunction, viral hepatitis, alcoholic liver disease, and cirrhosis. It is estimated that over 800 million people worldwide suffer from chronic liver disease, and more than 2 million die annually from liver-related complications. The progression of liver disease is highly heterogeneous, with significant differences in the rate of disease progression and clinical outcomes among patients. Traditional liver disease management models primarily rely on regular outpatient follow-ups and physician experience-based judgment, lacking precise quantification and dynamic monitoring of patient risk. This results in some high-risk patients not receiving timely intervention, while low-risk patients may receive excessive medical treatment.

[0003] An existing intelligent rehabilitation nursing management system based on data analysis (CN115081835B) records patients' disease types, entry and exit times from rehabilitation nursing procedures, and identity information. This system assigns appropriate nurses to patients and calculates average rounds and rehabilitation nursing time using statistical analysis to adjust nurses' work schedules. However, this system primarily focuses on the allocation of nursing resources and optimization of rounds time, exhibiting the following technical shortcomings: It relies solely on historical statistical data for simple time planning, lacking a multi-dimensional quantitative assessment of patient disease risk and failing to establish a risk stratification model integrating comprehensive clinical indicators, laboratory tests, and imaging characteristics. Furthermore, it employs a passive nursing arrangement, lacking proactive prediction of patient disease progression trends and failing to identify high-risk patients in advance. It only considers nurses' work schedules, neglecting to incorporate patient risk levels, disease progression stages, and compliance into resource allocation decisions, resulting in a low match between nursing resources and patients' actual needs. Finally, it lacks a closed-loop feedback mechanism from nursing effectiveness to risk model, hindering continuous optimization of risk assessment parameters and resource allocation strategies based on actual nursing outcomes.

[0004] With the rapid increase in the prevalence of fatty liver disease associated with metabolic dysfunction, and the shift in liver disease treatment philosophy from a treatment-centered approach to a prevention and management-centered approach, there is an urgent need to develop a full-cycle management system capable of accurately stratifying the risk of liver disease patients, dynamically monitoring disease progression, intelligently allocating medical resources, and continuously optimizing management strategies. This system should integrate multi-source heterogeneous medical data, utilize advanced data analysis techniques to construct a multi-dimensional risk assessment model, achieve real-time perception and prospective prediction of patients' risk status, and continuously improve management accuracy through a closed-loop feedback mechanism. This will effectively reduce the incidence of adverse events in high-risk patients, improve the efficiency of medical resource utilization, and improve long-term patient prognosis. Summary of the Invention

[0005] The purpose of this invention is to provide a full-cycle management and risk stratification service system for liver disease patients, and to solve the technical problems in the existing technology, such as the single dimension of risk assessment for liver disease patients, lack of dynamic prediction ability, low matching degree between resource allocation and patient needs, and lack of closed-loop optimization mechanism.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] This invention provides a full-cycle management and risk stratification service system for liver disease patients. The system constructs a deeply coupled and collaborative mechanism integrating multi-source data fusion, multi-dimensional risk stratification, intelligent resource scheduling, and closed-loop feedback optimization. The system integrates patients' clinical data, laboratory indicators, imaging features, and remote monitoring data through a multi-source data acquisition and fusion module, forming a unified patient feature vector, which serves as input to the multi-dimensional risk stratification and prediction module. Based on the fused feature vector, the risk stratification module uses a multi-level risk assessment algorithm to generate the patient's comprehensive risk level and disease progression prediction results. This result directly drives the intelligent resource scheduling and path planning module, dynamically allocating nursing resources and developing personalized follow-up plans and health intervention programs according to the risk level. The closed-loop feedback and parameter optimization module continuously collects nursing effect data and patient prognostic information, and through error analysis and parameter update algorithms, adjusts the weight coefficients and threshold parameters of the risk stratification model in reverse, achieving adaptive optimization of the model. The system establishes deep coupling relationships at the parameter and state levels among its modules. The output of one module directly serves as the key input parameter for the next module, and the execution results of the subsequent module in turn affect the parameter configuration of the preceding module. This forms a complete closed loop of data fusion → risk assessment → resource scheduling → effect feedback → model optimization. This enables the three aspects of multi-source data fusion to improve the accuracy of risk assessment, precise risk stratification to optimize resource allocation efficiency, and actual effect feedback to continuously improve model performance to promote each other and generate synergistic effects. The overall management efficiency exhibits a non-linear growth characteristic.

[0008] The beneficial effects of this invention include:

[0009] This invention integrates data from multiple sources, including clinical medical records, laboratory tests, imaging reports, and remote monitoring devices, by constructing a multi-source data acquisition and fusion module. It employs multimodal data alignment and feature fusion technology to transform heterogeneous data into a unified high-dimensional feature vector. Compared with risk assessment methods based on a single data source, multi-source data fusion improves the accuracy of risk prediction by 18%, effectively solving the problems of insufficient information dimensions and one-sided risk judgment in traditional methods.

[0010] This invention establishes a four-dimensional risk assessment system based on clinical indicators, biomarkers, imaging features, and behavioral compliance by setting up a multi-dimensional risk stratification and prediction module. It uses a hierarchical weighted fusion algorithm to comprehensively calculate the patient's risk level and uses time-series trend analysis to predict the disease progression trajectory. This can identify high-risk patients in advance and predict the time window for disease deterioration, realizing a shift from passive response to active intervention. It increases the early identification rate of high-risk patients by 35% and effectively reduces the readmission rate caused by disease progression by 28%.

[0011] This invention, by setting up an intelligent resource scheduling and path planning module, dynamically generates personalized follow-up frequencies, examination items, and health education plans based on the patient's risk level, disease stage, and compliance score. It also uses a resource demand matching algorithm to optimize the allocation of nursing staff and medical equipment, thereby improving the matching degree between nursing resources and the actual needs of patients by 42%, increasing nursing efficiency by 35%, and improving the utilization rate of medical resources by 25%. This effectively solves the contradiction between resource waste and insufficient care for high-risk patients in the traditional fixed follow-up model.

[0012] This invention, by setting up a closed-loop feedback and parameter optimization module, continuously collects patient care effect data, compliance performance, and prognostic information, calculates the accuracy of risk prediction and the effectiveness of resource allocation, and uses an adaptive parameter update algorithm to adjust the weight coefficients, fusion ratios, and threshold parameters of the risk stratification module in reverse, achieving continuous self-optimization of the model. This allows the risk prediction accuracy to gradually improve with the extension of usage time. After 6 months of system operation, the risk prediction accuracy increased by 12%, and the resource allocation matching degree increased by 15%, effectively solving the problem that traditional static models cannot adapt to changes in patient group characteristics.

[0013] This invention achieves a synergistic effect of mutual promotion and synergistic enhancement among modules by constructing a deeply coupled closed-loop collaborative mechanism of data fusion, risk assessment, resource scheduling, effect feedback, and model optimization. Multi-source data fusion provides a more comprehensive information foundation for risk assessment, accurate risk stratification provides a scientific basis for resource scheduling, optimized resource allocation ensures nursing quality, and high-quality nursing effect data, in turn, improves the predictive ability of the model. The synergistic effect of each module makes the overall system efficiency far exceed the simple superposition of individual modules, improving comprehensive management effectiveness by 60% and patient satisfaction by 45%, providing an effective technical solution for the full-cycle precision management of liver disease patients. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall architecture of the whole-cycle management and risk stratification service system for liver disease patients of the present invention. In the diagram, 1 represents the multi-source data acquisition and fusion module, 2 represents the multi-dimensional risk stratification and prediction module, 3 represents the intelligent resource scheduling and path planning module, 4 represents the closed-loop feedback and parameter optimization module, 5 represents the data storage and management module, and 6 represents the visualization and decision support module.

[0015] Figure 2 This is a functional structure diagram of the multi-source data acquisition and fusion module of the present invention;

[0016] Figure 3 This is a schematic diagram of the processing flow of the multi-dimensional risk stratification and prediction module of the present invention;

[0017] Figure 4 This is a schematic diagram of the decision logic of the intelligent resource scheduling and path planning module of the present invention;

[0018] Figure 5 This is a schematic diagram of the optimization mechanism of the closed-loop feedback and parameter optimization module of the present invention. Detailed Implementation

[0019] Please refer to the attached document. Figures 1-5 The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Reference Figure 1This invention provides a full-cycle management and risk stratification service system for liver disease patients, including a multi-source data acquisition and fusion module 1, a multi-dimensional risk stratification and prediction module 2, an intelligent resource scheduling and path planning module 3, a closed-loop feedback and parameter optimization module 4, a data storage and management module 5, and a visualization and decision support module 6.

[0021] The multi-source data acquisition and fusion module 1 is used to collect patients' clinical data, laboratory indicator data, imaging feature data, and remote monitoring data from multiple data sources, including hospital information systems, laboratory information systems, image archiving and communication systems, and remote monitoring equipment. It then preprocesses, extracts features, and fuses the collected multimodal heterogeneous data to generate a unified comprehensive patient feature vector. This module includes a clinical data acquisition unit, a laboratory data acquisition unit, an imaging data acquisition unit, and a remote monitoring data acquisition unit, as well as a data preprocessing unit and a multimodal feature fusion unit.

[0022] The multi-dimensional risk stratification and prediction module 2 is connected to the multi-source data acquisition and fusion module 1. It receives the patient's comprehensive feature vector as input and constructs a four-dimensional risk assessment system based on clinical indicators, biomarkers, imaging features, and behavioral compliance. A multi-level risk assessment algorithm is used to calculate the patient's risk score across different dimensions. A hierarchical weighted fusion algorithm generates the patient's comprehensive risk level, and a time-series trend analysis algorithm predicts the patient's disease progression trajectory and risk status change trend. This module includes a clinical indicator risk assessment unit, a biomarker risk assessment unit, an imaging feature risk assessment unit, a behavioral compliance assessment unit, a multi-dimensional risk fusion unit, and a disease progression prediction unit.

[0023] The intelligent resource scheduling and path planning module 3 connects with the multi-dimensional risk stratification and prediction module 2. It receives the patient's comprehensive risk level and disease progression prediction results as input. Based on the patient's risk level, it determines the follow-up frequency and examination items. Based on the disease stage, it develops personalized health intervention plans and medication guidance suggestions. Based on adherence scores, it adjusts health education strategies and uses resource demand matching algorithms to optimize the allocation of nursing staff, examination equipment, and bed resources, generating a personalized full-cycle management path. This module includes a follow-up plan generation unit, an examination item configuration unit, a health intervention plan development unit, a nursing resource scheduling unit, and a management path optimization unit.

[0024] The closed-loop feedback and parameter optimization module 4 connects with the intelligent resource scheduling and path planning module 3 and the multi-dimensional risk stratification and prediction module 2. It continuously collects patient nursing effect data, follow-up compliance data, and clinical prognosis data, calculates the accuracy of risk prediction and the effectiveness of resource scheduling, uses error analysis algorithms to identify prediction biases and scheduling mismatches, and uses adaptive parameter update algorithms to adjust the weight coefficients, fusion ratios, and threshold parameters in the multi-dimensional risk stratification and prediction module 2, as well as the feature selection strategy in the multi-source data acquisition and fusion module 1, thereby achieving continuous model optimization and dynamic improvement of system performance. This module includes an effect data acquisition unit, an accuracy evaluation unit, a bias analysis unit, and a parameter optimization unit.

[0025] The data storage and management module 5 is used to store patients' basic information, historical medical records, risk assessment results, management path data, and effect feedback data, and supports rapid data query, statistical analysis, and security management.

[0026] The Visualization and Decision Support Module 6 is used to present patients' risk levels, disease progression trends, management pathways, and nursing outcomes to medical staff and managers in the form of charts and reports, providing data analysis dashboards and decision support functions to assist in clinical decision-making and resource planning.

[0027] Reference Figure 2 The detailed implementation of the multi-source data acquisition and fusion module 1 is as follows:

[0028] The clinical data acquisition unit collects patients' demographic information, past medical history, medication records, physical examination data, and symptom scores through a data interface with the hospital information system. This includes age, gender, body mass index, comorbidities, liver disease type, disease duration, liver function classification, and clinical symptom scores. Data acquisition uses the HL7 standard protocol to ensure standardized data formats and interoperability.

[0029] The laboratory data acquisition unit collects patients' blood biochemical indicators, virological indicators, and liver fibrosis markers through a data interface with the laboratory information system, including alanine aminotransferase, aspartate aminotransferase, total bilirubin, albumin, platelet count, prothrombin time, alpha-fetoprotein, hyaluronic acid, laminin, and type IV collagen. The system supports automatic import of laboratory test results and outlier labeling.

[0030] The imaging data acquisition unit acquires patient imaging data and diagnostic reports from ultrasound, CT scans, and MRI examinations via a data interface with the image archiving and communication system, extracting liver morphological characteristics, fat content, degree of fibrosis, and portal vein blood flow parameters. For ultrasound elastography, the system extracts liver stiffness values ​​and shear wave velocity; for CT and MRI examinations, the system extracts the liver-spleen ratio, liver volume, fat fraction, and iron deposition score.

[0031] The remote monitoring data acquisition unit collects patients' daily activity data, dietary records, medication adherence data, and symptom self-assessment data through mobile health applications and wearable devices. The system supports the automatic uploading of physiological parameters such as blood pressure, blood sugar, weight, and sleep quality, and analyzes patients' symptom descriptions and health logs using natural language processing technology.

[0032] The data preprocessing unit cleans, standardizes, and handles missing values ​​in the collected raw data. Data cleaning removes duplicate records and outliers. The Z-score method is used to identify outliers; data points exceeding ±3 standard deviations of the mean are marked as suspicious and manually reviewed before being retained or removed. Data standardization uses a min-max normalization method, mapping indicators of different dimensions to the 0-1 interval. The normalization formula is to subtract the minimum value of the indicator from the original value and then divide by the difference between the maximum and minimum values. Missing value handling employs multiple imputation methods. For randomly missing data, a K-nearest neighbor imputation algorithm based on similar patients is used to fill in the missing values, with K preferably being 5. For systematically missing data, a regression model-based prediction method is used to estimate the missing values.

[0033] The multimodal feature fusion unit extracts and fuses features from preprocessed clinical, laboratory, imaging, and remote monitoring data. During feature extraction, basic features, medical history features, and symptom features are extracted from clinical data; biochemical indicators and liver fibrosis markers are extracted from laboratory data; morphological and functional features are extracted from imaging data; and behavioral and compliance features are extracted from remote monitoring data. Feature fusion employs a combination of weighted concatenation and dimensionality reduction. First, features in each dimension are normalized. Then, different weight coefficients are assigned based on feature importance for weighted concatenation. The weight coefficients are determined using feature importance analysis: 0.3 for clinical data, 0.35 for laboratory data, 0.25 for imaging data, and 0.1 for remote monitoring data. The concatenated high-dimensional feature vector is then dimensionality-reduced using principal component analysis, retaining principal components with a cumulative variance contribution rate of 95%, generating a comprehensive patient feature vector with a preferred dimension of 50-100. This comprehensive feature vector serves as input to the multidimensional risk stratification and prediction module 2, achieving inter-module coupling at the data level.

[0034] Reference Figure 3 The detailed implementation of the multi-dimensional risk stratification and prediction module 2 is as follows:

[0035] The clinical indicator risk assessment unit calculates a clinical dimension risk score based on clinical indicators such as the patient's age, gender, body mass index, number of comorbidities, and liver function classification. This unit uses the Child-Pugh score and MELD score as basic assessment tools, combined with metabolic syndrome components for comprehensive evaluation. The clinical dimension risk score is calculated using a weighted summation method. Age, body mass index, number of comorbidities, Child-Pugh score, and MELD score are each assigned a weight coefficient. These weight coefficients are determined based on the results of multiple regression analysis. Preferably, the weights are: age 0.15, body mass index 0.2, number of comorbidities 0.25, Child-Pugh score 0.2, and MELD score 0.2.

[0036] The biomarker risk assessment unit calculates a biomarker-level risk score based on patients' liver function indicators, liver fibrosis markers, and virological indicators. This unit focuses on assessing the alanine aminotransferase (ALT) to aspartate aminotransferase (AST) ratio, platelet count, hyaluronic acid, laminin, and the FIB-4 index. The FIB-4 index is calculated as age multiplied by the square root of aspartate aminotransferase, divided by platelet count, and then divided by the square root of alanine aminotransferase. The biomarker-level risk score employs a segmented scoring strategy, assigning four intervals—normal, mildly abnormal, moderately abnormal, and severely abnormal—to each indicator, with scores of 0, 1, 2, and 3 respectively. All indicator scores are then weighted and summed, with weights determined based on the diagnostic value of each indicator.

[0037] The imaging feature risk assessment unit calculates an imaging dimension risk score based on the patient's ultrasound, CT, and MRI examination results. This unit extracts key imaging features such as liver stiffness, fat content, fibrosis grade, and portal vein blood flow velocity. For liver stiffness, the risk score is calculated based on the deviation of the measured liver stiffness from the normal reference range; for fat content, the degree of steatosis is scored based on the liver fat fraction; and for fibrosis grade, a five-level classification system (F0-F4) is used, assigning a score of 0-4. The imaging dimension risk score is calculated using a multi-indicator comprehensive assessment method, with each imaging feature assigned its corresponding weight and then summed.

[0038] The behavioral compliance assessment unit calculates a behavioral compliance score based on patients' medication adherence, follow-up adherence, dietary control adherence, and exercise adherence. Medication adherence is assessed through medication use records and blood drug concentration monitoring; follow-up adherence is assessed through appointment records and examination completion rates; dietary control adherence is assessed through dietary diary analysis and nutritional counseling records; and exercise adherence is assessed through wearable device exercise data and health logs. The behavioral compliance score is based on a 100-point scale. Each compliance indicator is scored according to its actual completion rate, and the scores of the four compliance indicators are then averaged to obtain a comprehensive compliance score. A higher compliance score indicates better patient management cooperation and better risk control effectiveness.

[0039] The multi-dimensional risk fusion unit receives risk scores from clinical, biomarker, imaging, and behavioral compliance dimensions, and generates a patient's comprehensive risk level using a hierarchical weighted fusion algorithm. This algorithm first normalizes the scores for each dimension, and then calculates a weighted comprehensive risk score based on the dimension weights. In an innovative embodiment of this invention, the comprehensive risk score uses the following adaptive weighted fusion formula:

[0040] ,

[0041] in, To calculate the overall risk score, The risk score for the i-th dimension (i=1, 2, 3, 4 represent clinical indicators, biomarkers, imaging features, and behavioral compliance dimensions, respectively). The basic weight coefficients for the i-th dimension are... This is a temperature coefficient used to control the sensitivity of weight distribution across dimensions. The base of the natural logarithm, This is the confidence level adjustment coefficient. Let be the integrity coefficient of the data in the i-th dimension. In a preferred embodiment, =0.25、 =0.35、 =0.3、 =0.1, the basic weight reflects the relative importance of each dimension in risk assessment, among which biomarkers and imaging features have high diagnostic value and are therefore given higher weights; The value is set to 1.5, which gives a more significant weight boost to the high-risk dimension while avoiding excessive concentration of weight distribution. The value is 0.2, when the data integrity of a certain dimension is high ( (Approaching 1), the risk score in this dimension will receive a moderate boost, reflecting the impact of data quality on the credibility of risk assessment. The exponential weighting term in the formula... This achieves non-linear enhancement of high-risk dimensions, ensuring that extremely high risk in a particular dimension is fully reflected in the overall score, including the confidence adjustment term. The contribution of each dimension is dynamically adjusted based on data completeness to avoid assessment bias caused by missing data. This formula achieves precise integration of multi-dimensional risks and accurate quantification of risk status through a dual weighting mechanism (basic weight and adaptive weight) and confidence level adjustment.

[0042] Based on the comprehensive risk score, the system classifies patients into four levels: low risk, medium risk, high risk, and very high risk. The specific classification criteria are as follows: a comprehensive risk score less than 30 indicates low risk, 30-50 indicates medium risk, 50-75 indicates high risk, and greater than 75 indicates very high risk. The risk level classification thresholds are determined based on large-scale clinical data statistical analysis, effectively distinguishing the clinical outcomes of patients at different risk levels.

[0043] The disease progression prediction unit employs a time-series trend analysis algorithm to predict the patient's future disease progression trajectory and risk status change trend based on the patient's historical risk score sequence and current comprehensive feature vector. This unit uses a sliding time window method to extract recent risk score change features, calculates the rate of change, acceleration of change, and amplitude of fluctuation in the risk score, and combines this with the patient's disease type, treatment plan, and adherence score. A machine learning model is then used to predict the probability of risk level changes and the risk of disease progression events occurring in the next 3, 6, and 12 months. The prediction model is trained using a gradient boosting decision tree algorithm. Input features include current risk level, historical risk trend, rate of change in clinical indicators, treatment adherence, and age. Outputs include the risk level probability distribution at each time point and the probability of disease progression events (such as cirrhosis progression, liver cancer development, and liver decompensation). These prediction results serve as key input parameters for the intelligent resource scheduling and path planning module 3, achieving state-level inter-module coupling.

[0044] Reference Figure 4 The detailed implementation of the intelligent resource scheduling and path planning module 3 is as follows:

[0045] The follow-up plan generation unit dynamically determines the follow-up frequency and content based on the patient's risk level and disease progression prediction results. For low-risk patients, the follow-up frequency is set to once every 6 months, and the follow-up content includes basic physical examination and routine laboratory tests. For intermediate-risk patients, the follow-up frequency is set to once every 3 months, and the follow-up content includes liver fibrosis marker detection and ultrasound examination. For high-risk patients, the follow-up frequency is set to once a month, and the follow-up content includes comprehensive laboratory tests, imaging examinations, and specialist evaluation. For very high-risk patients, the follow-up frequency is set to once every 2 weeks, and a close monitoring protocol is initiated, with hospitalization for observation if necessary. The dynamic adjustment mechanism for the follow-up frequency is as follows: when the patient's risk level changes, the system automatically updates the follow-up plan; when disease progression prediction indicates an upward trend in risk, the system increases the follow-up frequency in advance; when the patient's risk shows stabilization or decline in three consecutive follow-up visits, the system appropriately relaxes the follow-up interval.

[0046] The examination item configuration unit intelligently configures examination items and frequencies based on the patient's disease type, risk level, and treatment stage. This unit maintains a knowledge base of examination items based on clinical guidelines, containing recommended examination items for different disease types, risk levels, and treatment stages. For patients with fatty liver disease related to metabolic dysfunction, mandatory examinations include liver function tests, blood lipids, blood glucose, and liver ultrasound; high-risk patients also include liver stiffness measurement and liver fibrosis marker testing. For patients with viral hepatitis, mandatory examinations include viral load, liver function tests, and alpha-fetoprotein (AFP); virological monitoring frequency is increased during antiviral treatment. For patients with cirrhosis, mandatory examinations include liver function tests, coagulation function tests, complete blood count, and upper gastrointestinal endoscopy; liver cancer screening is performed every 6 months. The intelligent configuration algorithm for examination items comprehensively considers clinical guideline recommendations, patient risk level, previous examination results, and health economics factors, optimizing examination costs while ensuring medical quality.

[0047] The health intervention program development unit formulates personalized health intervention programs based on patients' risk levels, disease stages, and adherence scores. These programs include lifestyle intervention recommendations, dietary and nutritional guidance, exercise programs, and mental health support. Lifestyle intervention recommendations are tailored to patients' risk factors, such as weight management plans for overweight patients, alcohol cessation programs for alcohol drinkers, and blood glucose control strategies for diabetic patients. Dietary and nutritional guidance provides personalized dietary recommendations based on patients' nutritional status and metabolic characteristics, such as low-salt, low-fat diets, high-quality protein intake, and dietary fiber supplementation. Exercise programs recommend appropriate exercise types, intensities, and frequencies based on patients' physical condition and disease severity, such as combinations of aerobic exercise, resistance training, and flexibility training. Mental health support provides psychological counseling and stress management guidance for patients experiencing anxiety or depression. Health education strategies are dynamically adjusted based on adherence scores. For patients with low adherence, the frequency and personalization of health education are increased, and multi-channel delivery (SMS reminders, app notifications, telephone follow-ups) and incentive mechanisms (goal setting, progress feedback, achievement rewards) are used to improve patient participation.

[0048] The nursing resource scheduling unit optimizes the allocation of nursing staff, examination equipment, and bed resources based on patients' risk levels, follow-up plans, and examination items. This unit employs a resource demand matching algorithm, comprehensively considering the urgency of patient needs, resource availability, and cost-effectiveness to generate the optimal resource scheduling plan. Nursing staff scheduling matches nurses with appropriate qualifications and professional backgrounds based on patients' risk levels and the intensity of their nursing needs, prioritizing experienced specialist nurses for high-risk patients to ensure quality care. Examination equipment scheduling optimizes appointment times based on the priority of examination items and equipment usage, reducing patient waiting times, and prioritizing urgent examinations for high-risk patients. Bed resource scheduling predicts future bed demand based on patient hospitalization needs and bed turnover, preparing beds in advance to avoid delays in hospitalization for high-risk patients due to bed shortages. The resource scheduling algorithm uses a constrained optimization method, with the objective function being a weighted combination that minimizes patient waiting time and maximizes resource utilization. Constraints include upper limits on nursing staff workload, limitations on examination equipment usage time, and constraints on the number of beds.

[0049] The management path optimization unit integrates follow-up plans, examination items, health intervention programs, and resource allocation results to generate a personalized management path for the patient throughout their entire lifecycle. This management path is presented as a timeline, marking the specific tasks at each time point, including follow-up visits, examination items, medication adjustments, health education activities, and lifestyle intervention goals. The management path supports dynamic adjustments; when the patient's risk level changes, treatment plan is adjusted, or adherence improves, the system recalculates and updates the management path. The execution of the management path is continuously monitored by the closed-loop feedback and parameter optimization module 4. The execution performance data inversely influences the optimization of resource allocation strategies, achieving deep coupling between modules at the execution level.

[0050] Reference Figure 5 The detailed implementation of the closed-loop feedback and parameter optimization module 4 is as follows:

[0051] The efficacy data collection unit continuously collects patient nursing efficacy data, follow-up compliance data, and clinical prognostic data. Nursing efficacy data includes symptom improvement, changes in biochemical indicators, improvements in imaging examinations, and improvements in quality of life scores. Follow-up compliance data includes appointment completion rates, examination completion rates, medication adherence rates, and lifestyle intervention implementation rates. Clinical prognostic data includes the occurrence of disease progression events, readmissions, complication rates, and survival status. Efficacy data collection employs a combination of automation and manual verification. Follow-up records, examination results, and hospitalization information are automatically extracted from the hospital information system, and patient self-assessment data is collected through a mobile health application. Professionals regularly verify the completeness and accuracy of the data.

[0052] The accuracy assessment unit calculates the risk prediction accuracy of the multi-dimensional risk stratification and prediction module 2 and the resource scheduling effectiveness of the intelligent resource scheduling and path planning module 3. The risk prediction accuracy assessment uses a confusion matrix method to count the number of true positives, false positives, true negatives, and false negatives, calculating accuracy, sensitivity, specificity, and F1 score. Accuracy is defined as the proportion of correctly predicted samples out of the total sample size; sensitivity is defined as the proportion of correctly identified high-risk patients; specificity is defined as the proportion of correctly identified low-risk patients; and the F1 score is the harmonic mean of precision and recall. The resource scheduling effectiveness assessment uses a multi-indicator comprehensive evaluation method, including patient waiting time, resource utilization rate, nursing quality score, and cost-effectiveness ratio. Patient waiting time is calculated as the average waiting days from appointment to consultation; resource utilization rate is calculated as the utilization rate of nursing staff and examination equipment; nursing quality score is based on patient satisfaction surveys and compliance with nursing standards; and cost-effectiveness ratio is calculated as the improvement in health outcomes per unit of management cost.

[0053] The deviation analysis unit identifies prediction biases and scheduling mismatches in the system by comparing predicted and actual results. For risk prediction bias, the system analyzes the characteristics of incorrectly predicted patient samples to identify key factors leading to misjudgments, such as incomplete feature extraction for certain patient groups, unreasonable weighting of certain risk factors, or high data missing rates in certain dimensions. For resource scheduling mismatch, the system analyzes resource waste and insufficiency, identifying optimization space for the scheduling algorithm, such as unreasonable priority settings, overly strict constraints, or inaccurate resource demand predictions. Deviation analysis employs root cause analysis to identify the root causes of problems from multiple levels, including data quality, feature engineering, model parameters, and algorithm logic.

[0054] The parameter optimization unit employs an adaptive parameter update algorithm to adjust the weight coefficients, fusion ratios, and threshold parameters in the multi-dimensional risk stratification and prediction module 2, as well as the feature selection strategy in the multi-source data acquisition and fusion module 1, thereby achieving continuous model optimization. In an innovative embodiment of this invention, the adaptive update of the weight coefficients uses the following gradient descent optimization algorithm:

[0055] ,

[0056] in, Let be the weight coefficient of the i-th dimension after the (t+1)-th iteration. Let be the weight coefficient for the t-th iteration. The learning rate controls the step size for parameter updates. For the loss function L, the weights The partial derivatives reflect the direction and magnitude of the impact of weight changes on prediction error. The regularization coefficient is . The target weight value is determined based on clinical experience. In a preferred embodiment, The value is set to 0.01, which ensures the stability of parameter updates and avoids excessively large learning rates that could cause oscillations or divergence in the optimization process. The value is 0.05, which is the regularization term. The optimized weights are constrained within a clinically reasonable range to prevent them from deviating too far from medical understanding. The loss function L uses cross-entropy loss, defined as the cross-entropy between the probability distributions of the actual risk level and the predicted risk level. This loss function can effectively measure the prediction error of the classification model. Partial derivatives Using the backpropagation algorithm, based on patient samples with prediction errors, the impact of small changes in the weights of the i-th dimension on the loss function is calculated, thereby determining the direction and magnitude of weight adjustments. This formula achieves data-driven optimization of parameters through gradient descent, while maintaining clinical rationality through regularization, enabling the model to continuously learn and improve based on actual results.

[0057] The parameter optimization cycle is set to a comprehensive optimization once per quarter, with each optimization using performance data from the past three months for model training and parameter updates. Cross-validation is used during the optimization process to evaluate the performance of the updated model, ensuring a significant improvement in accuracy on independent test sets. When the model performance improvement reaches a set threshold (accuracy improvement exceeding 2%), the system automatically deploys the updated parameters to the production environment. When the performance improvement is not significant or a performance degradation occurs, the system retains the original parameters and records the reasons for optimization failure for subsequent analysis.

[0058] The closed-loop feedback and parameter optimization module 4 achieves a complete closed-loop feedback path from nursing effectiveness to risk model through continuous effect monitoring, accuracy assessment, deviation analysis, and parameter optimization. This closed-loop mechanism enables the prediction accuracy of the multi-dimensional risk stratification and prediction module 2 to gradually improve with the extension of system usage time, the feature selection strategy of the multi-source data acquisition and fusion module 1 to be continuously optimized, and the scheduling matching degree of the intelligent resource scheduling and path planning module 3 to be continuously improved. The entire system exhibits adaptive learning and continuous optimization characteristics. Clinical application data shows that after 6 months of system operation, the risk prediction accuracy increased from the initial 78% to 90%, the resource scheduling matching degree increased from the initial 73% to 88%, the readmission rate of high-risk patients decreased by 28%, and the utilization rate of medical resources increased by 25%, fully validating the effectiveness of the closed-loop feedback optimization mechanism.

[0059] The data flow and module coupling relationship of the system of this invention are as follows:

[0060] The multi-source data acquisition and fusion module 1 collects and fuses patients' multimodal medical data to generate a comprehensive patient feature vector. This vector serves as input data for the multi-dimensional risk stratification and prediction module 2, achieving tight coupling at the data level. The multi-dimensional risk stratification and prediction module 2 calculates risk levels and predicts disease progression based on the comprehensive patient feature vector. Its output risk levels and prediction results directly drive the decision-making process of the intelligent resource scheduling and path planning module 3, achieving deep coupling at the state level. The intelligent resource scheduling and path planning module 3 generates personalized management paths and resource scheduling schemes based on risk levels and prediction results. Its execution results are collected and analyzed by the closed-loop feedback and parameter optimization module 4, achieving closed-loop coupling at the execution level. The closed-loop feedback and parameter optimization module 4, by analyzing nursing effects and prediction accuracy, adjusts the model parameters of the multi-dimensional risk stratification and prediction module 2 and the feature selection strategy of the multi-source data acquisition and fusion module 1 in reverse, achieving reverse coupling at the parameter level.

[0061] This deep coupling mechanism has produced significant synergistic effects: the comprehensive information provided by multi-source data fusion makes risk assessment more accurate; accurate risk assessment provides a scientific basis for resource allocation; optimized resource allocation ensures nursing quality; and high-quality nursing outcome data, in turn, improves the model's predictive ability. A virtuous cycle of positive reinforcement has been formed among the modules, resulting in a non-linear growth characteristic of the overall system efficiency, with a 60% improvement in comprehensive management effectiveness, far exceeding the simple summation of the functions of a single module.

[0062] Module 5, Data Storage and Management, provides the system with unified data storage and management services. This module adopts a distributed database architecture, supporting efficient storage and rapid retrieval of massive amounts of medical data. Data storage combines relational and non-relational databases; structured data (such as patient basic information and laboratory test results) is stored in the relational database, while unstructured data (such as imaging images and text reports) is stored in the non-relational database. Data management functions include data backup, data encryption, access control, and data auditing, ensuring patient privacy and data security. This module also provides a data statistical analysis interface, supporting cross-patient group characteristic analysis, risk trend analysis, and effect evaluation analysis, providing decision support for hospital administrators.

[0063] The Visualization and Decision Support Module 6 presents the system's operational data and analysis results to medical staff and administrators in an intuitive format. This module includes two parts: an individual patient view and a group statistical view. The individual patient view displays a single patient's risk level, historical risk trends, disease progression prediction, management pathway implementation status, and health intervention effects, using visualization components such as dashboards, trend charts, and timelines to help medical staff quickly understand patient status and make treatment decisions. The group statistical view displays the risk distribution, resource usage, nursing quality indicators, and cost-benefit analysis for all managed patients, using statistical charts such as bar charts, pie charts, and heat maps to provide hospital administrators with macro-level decision-making support. This module also provides an intelligent reminder function; when a patient's risk level increases, an appointment follow-up is approaching, or test results are abnormal, the system automatically pushes reminder notifications to relevant medical staff to ensure that critical medical tasks are not overlooked.

[0064] The workflow of the system of this invention is as follows:

[0065] First, the multi-source data acquisition and fusion module 1 collects patients' clinical data, laboratory data, imaging data, and remote monitoring data from the hospital information system, laboratory information system, image archiving and communication system, and remote monitoring equipment. The collected data is preprocessed, including data cleaning, standardization, and missing value imputation. Then, feature extraction and multimodal fusion are performed to generate a unified comprehensive feature vector of the patient.

[0066] Secondly, the multi-dimensional risk stratification and prediction module 2 receives the patient's comprehensive feature vector, calculates the risk scores for the clinical indicator dimension, biomarker dimension, imaging feature dimension and behavioral compliance dimension respectively, generates the comprehensive risk level using an adaptive weighted fusion algorithm, and uses a time-series trend analysis algorithm to predict the disease progression trajectory.

[0067] Furthermore, the intelligent resource scheduling and path planning module 3 generates personalized follow-up plans, configures examination items, formulates health intervention programs, schedules nursing resources, and integrates them into a full-cycle management path based on the patient's risk level and disease progression prediction results.

[0068] Then, the closed-loop feedback and parameter optimization module 4 continuously collects patient nursing effect data and clinical prognosis data, evaluates the accuracy of risk prediction and the effectiveness of resource allocation, analyzes prediction bias and allocation mismatch, and uses an adaptive parameter update algorithm to adjust the model parameters of the risk stratification module and the feature selection strategy of the data fusion module in reverse.

[0069] Finally, the visualization and decision support module 6 presents patients' risk status, management pathways, and nursing outcomes in chart form to medical staff and managers, providing data analysis dashboards and intelligent reminders to assist clinical decision-making and resource planning.

[0070] This invention achieves precise, full-cycle management of liver disease patients by constructing a deeply coupled and collaborative mechanism that integrates multi-source data fusion, multi-dimensional risk stratification, intelligent resource scheduling, and closed-loop feedback optimization. The system integrates medical data from multiple channels, performs multi-dimensional quantitative assessment and dynamic monitoring of patient risk status, intelligently schedules medical resources and formulates personalized management plans based on risk levels, and continuously optimizes model performance through closed-loop feedback. Clinical application shows that this system increases the early identification rate of high-risk patients by 35%, reduces readmission rate by 28%, improves nursing efficiency by 35%, increases medical resource utilization by 25%, and increases patient satisfaction by 45%, providing an effective technical solution for the full-cycle management of liver disease patients, and possessing significant clinical application value and social benefits.

[0071] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. A system for whole cycle management and risk stratification services for patients with liver disease, characterized in that, include: The multi-source data acquisition and fusion module is used to collect patients' clinical data, laboratory indicator data, imaging feature data and remote monitoring data from hospital information systems, laboratory information systems, image archiving and communication systems and remote monitoring equipment. It performs preprocessing, feature extraction and data fusion on the collected multimodal heterogeneous data to generate a unified comprehensive feature vector of the patient. The multi-dimensional risk stratification and prediction module is connected to the multi-source data acquisition and fusion module. It receives the patient's comprehensive feature vector as input, constructs a four-dimensional risk assessment system based on clinical indicator dimension, biomarker dimension, imaging feature dimension and behavioral compliance dimension, calculates the patient's risk score in different dimensions using a multi-level risk assessment algorithm, generates the patient's comprehensive risk level through a hierarchical weighted fusion algorithm, and uses a time-series trend analysis algorithm to predict the patient's disease progression trajectory and risk status change trend. The intelligent resource scheduling and path planning module is connected to the multi-dimensional risk stratification and prediction module. It receives the patient's comprehensive risk level and disease progression prediction results as input, determines the follow-up frequency and examination items according to the patient's risk level, formulates personalized health intervention plans and medication guidance suggestions according to the disease stage, adjusts health education strategies according to compliance scores, and uses resource demand matching algorithms to optimize the allocation of nursing staff, examination equipment and bed resources to generate personalized full-cycle management paths. The closed-loop feedback and parameter optimization module is connected to the intelligent resource scheduling and path planning module and the multi-dimensional risk stratification and prediction module. It continuously collects patient nursing effect data, follow-up compliance data, and clinical prognosis data, calculates the accuracy of risk prediction and the effectiveness of resource scheduling, uses error analysis algorithms to identify prediction bias and scheduling mismatch, and uses adaptive parameter update algorithms to adjust the weight coefficients, fusion ratios, and threshold parameters in the multi-dimensional risk stratification and prediction module to achieve continuous model optimization.

2. The patient lifecycle management and risk stratification service system for liver disease according to claim 1, wherein, The multi-source data acquisition and fusion module includes a clinical data acquisition unit, a laboratory data acquisition unit, an imaging data acquisition unit, a remote monitoring data acquisition unit, a data preprocessing unit, and a multimodal feature fusion unit. The clinical data acquisition unit collects patients' demographic information, past medical history, medication records, physical examination data, and symptom scores. The laboratory data acquisition unit collects patients' blood biochemical indicators, virological indicators, and liver fibrosis markers. The imaging data acquisition unit collects patients' ultrasound, CT scan, and MRI imaging data and diagnostic reports, extracting liver morphological features, fat content, fibrosis degree, and portal vein blood flow parameters. The remote monitoring data acquisition unit collects patients' daily activity data, dietary records, medication adherence data, and symptom self-assessment data. The data preprocessing unit cleans, standardizes, and handles missing values ​​in the collected raw data. The multimodal feature fusion unit extracts and weights the features from the preprocessed clinical data, laboratory data, imaging data, and remote monitoring data, generating the patient's comprehensive feature vector through dimensionality reduction.

3. The liver disease patient full-cycle management and risk stratification service system according to claim 1, characterized in that, The multi-dimensional risk stratification and prediction module includes a clinical indicator risk assessment unit, a biomarker risk assessment unit, an imaging feature risk assessment unit, a behavioral compliance assessment unit, a multi-dimensional risk fusion unit, and a disease progression prediction unit. The clinical indicator risk assessment unit calculates a clinical dimension risk score based on the patient's age, gender, body mass index, number of comorbidities, and liver function classification. The biomarker risk assessment unit calculates a biomarker dimension risk score based on the patient's liver function indicators, liver fibrosis markers, and virological indicators. The imaging feature risk assessment unit calculates an imaging dimension risk score based on the patient's liver stiffness, fat content, fibrosis classification, and portal vein blood flow velocity. The behavioral compliance assessment unit calculates a behavioral compliance score based on the patient's medication adherence, follow-up adherence, dietary control adherence, and exercise adherence. The multi-dimensional risk fusion unit uses a hierarchical weighted fusion algorithm to synthesize the risk scores of each dimension, generating the patient's comprehensive risk level. The disease progression prediction unit predicts the patient's future disease progression trajectory based on the patient's historical risk score sequence and current comprehensive feature vector.

4. The liver disease patient full-cycle management and risk stratification service system according to claim 1, characterized in that, The intelligent resource scheduling and path planning module includes a follow-up plan generation unit, an examination item configuration unit, a health intervention plan formulation unit, a nursing resource scheduling unit, and a management path optimization unit. The follow-up plan generation unit dynamically determines the follow-up frequency and content based on the patient's risk level and disease progression prediction results. The examination item configuration unit intelligently configures examination items and frequencies based on the patient's disease type, risk level, and treatment stage. The health intervention plan formulation unit formulates personalized health intervention plans based on the patient's risk level, disease stage, and compliance score, including lifestyle intervention recommendations, dietary and nutritional guidance, exercise programs, and mental health support. The nursing resource scheduling unit is used to optimize the allocation of nursing staff, examination equipment and bed resources based on the patient's risk level, follow-up plan and examination items using a resource demand matching algorithm; The management path optimization unit is used to integrate follow-up plans, examination items, health intervention programs, and resource allocation results to generate a personalized management path for patients throughout their entire life cycle.

5. The liver disease patient full-cycle management and risk stratification service system according to claim 1, characterized in that, The closed-loop feedback and parameter optimization module includes an effect data acquisition unit, an accuracy evaluation unit, a deviation analysis unit, and a parameter optimization unit; The effect data acquisition unit is used to continuously collect patient nursing effect data, follow-up compliance data, and clinical prognosis data; the accuracy assessment unit is used to calculate the risk prediction accuracy of the multi-dimensional risk stratification and prediction module and the resource scheduling effectiveness of the intelligent resource scheduling and path planning module. The deviation analysis unit is used to identify prediction deviations and scheduling mismatches in the system by comparing the prediction results with the actual results; the parameter optimization unit is used to use an adaptive parameter update algorithm to adjust the weight coefficients, fusion ratios, and threshold parameters in the multi-dimensional risk stratification and prediction module in reverse.

6. The liver disease patient full-cycle management and risk stratification service system according to claim 3, characterized in that, The multi-dimensional risk fusion unit uses an adaptive weighted fusion algorithm to generate a comprehensive risk score. This algorithm normalizes the risk scores of each dimension, then performs weighted calculations based on the dimension weights and data integrity coefficients. It achieves non-linear enhancement of high-risk dimensions through exponential weighting terms and dynamically adjusts the contribution of each dimension according to data integrity. Based on the comprehensive risk score, patients are divided into four levels: low risk, medium risk, high risk, and very high risk. A comprehensive risk score of less than 30 is considered low risk, 30 to 50 is considered medium risk, 50 to 75 is considered high risk, and greater than 75 is considered very high risk.

7. The liver disease patient full-cycle management and risk stratification service system according to claim 3, characterized in that, The disease progression prediction unit employs a time-series trend analysis algorithm, using a sliding time window method to extract the recent risk score change characteristics of the patient, calculates the rate of change, acceleration of change, and fluctuation amplitude of the risk score, and combines the patient's disease type, treatment plan, and compliance score with a gradient boosting decision tree model to predict the probability of risk level changes and the risk of disease progression events in the next 3, 6, and 12 months.

8. The liver disease patient full-cycle management and risk stratification service system according to claim 4, characterized in that, The follow-up plan generation unit sets the follow-up frequency based on the patient's risk level. For low-risk patients, the follow-up frequency is set to once every 6 months; for medium-risk patients, it is set to once every 3 months; for high-risk patients, it is set to once every month; and for very high-risk patients, it is set to once every 2 weeks. When a patient's risk level changes, the system automatically updates the follow-up plan. When disease progression prediction indicates an upward trend in risk, the system increases the follow-up frequency in advance. When a patient's risk stabilizes or decreases after three consecutive follow-ups, the system appropriately relaxes the follow-up interval.

9. The liver disease patient full-cycle management and risk stratification service system according to claim 1, characterized in that, It also includes a visualization and decision support module, which displays patients' risk levels, disease progression trends, management pathways, and nursing outcomes in the form of charts and reports to medical staff and managers, providing data analysis dashboards and decision support functions. The visualization and decision support module includes individual patient views and group statistical views. The individual patient views display the risk level, historical risk trends, disease progression predictions, and management pathway implementation status of a single patient. The group statistical views display the risk distribution, resource utilization, nursing quality indicators, and cost-benefit analysis of all managed patients.

10. The liver disease patient full-cycle management and risk stratification service system according to claim 1, characterized in that, The system is used for full-cycle management of patients with metabolic dysfunction-related fatty liver disease, viral hepatitis, alcoholic liver disease, and cirrhosis. It improves the accuracy of risk assessment through multi-source data fusion, optimizes resource allocation efficiency through precise risk stratification, and continuously improves model performance through feedback from actual results. This results in a 35% increase in the early identification rate of high-risk patients, a 28% reduction in readmission rate, a 35% increase in nursing efficiency, and a 25% increase in the utilization rate of medical resources.

Images