Method and system for predicting multiple diseases of old people

Abstract

The invention relates to the technical field of health management of old people, and discloses a method and a system for predicting multiple diseases of old people. The method comprises the following steps: acquiring multi-modal health monitoring data of a target old person in a preset time period, wherein the multi-modal health monitoring data comprises physiological index time sequence data, a medication record sequence and a daily activity ability evaluation result; and performing cross-modal correlation analysis on the multi-modal health monitoring data to generate a disease interaction characteristic matrix containing metabolic disease correlation degree, circulatory system disease coordination index and neurodegenerative disease progress rate. And performing hierarchical clustering processing on the feature matrix by adopting a dynamic weight distribution algorithm, and outputting potential common disease combinations of the target old people and a priority score of each common disease combination. And generating a personalized intervention strategy set including a drug interaction avoidance scheme, a rehabilitation training intensity adjustment scheme and a nutrition intake ratio scheme according to the priority score.

Description

Technical Field

[0001] This invention relates to the field of elderly health management technology, specifically to a method and system for predicting multiple comorbidities in the elderly. Background Technology

[0002] With the accelerating aging of the global population, the coexistence of multiple diseases has become a common health challenge for the elderly. Coexistence of multiple diseases refers to an individual simultaneously suffering from two or more chronic diseases, a situation particularly prevalent among the elderly. Complex pathophysiological mechanisms cause multiple chronic diseases to influence each other, forming complex comorbidity networks, such as common combinations like cardiovascular disease and chronic kidney disease, or diabetes and osteoporosis. Traditional single-disease treatment models struggle to cope with this complexity, often leading to fragmented treatment plans, increased risks associated with multiple medications, and inefficient use of medical resources.

[0003] Current clinical practice primarily relies on the independent diagnosis and experience-based judgment of specialists to identify comorbidity risks. This approach has significant limitations. First, the diagnostic process is delayed, typically intervening only after obvious clinical symptoms appear, missing the golden window for early prevention and management. Second, physician judgment is easily influenced by subjective experience, lacking quantitative analysis of the potential interaction mechanisms between multiple diseases. For example, the cumulative burden of different diseases on the same organ system, and the antagonistic or synergistic effects between drugs, are difficult to grasp intuitively. Furthermore, the health data of the elderly comes from diverse sources, including dynamic physiological parameters, medication adherence, and functional activity levels, among other multidimensional information. This data is often isolated in time and space, lacking effective methods for integrated analysis.

[0004] At the technical level, existing research has attempted to apply statistical models or basic machine learning methods for disease prediction. These methods often focus on the risk of a single disease or simple combinations of limited risk factors, failing to delve into the cross-disease associations hidden behind multimodal health data. For example, conventional methods may assess the risk of hypertension and diabetes separately, but rarely quantify the dynamic processes by which these two factors synergistically accelerate target organ damage. Furthermore, these methods typically employ fixed-weight models, which cannot adapt to the dynamic evolution of individual health conditions and are difficult to generate time-series-specific personalized intervention plans. The fragmentation of data processing drowns out potential early warning signals, failing to provide a systematic and forward-looking reference for clinical decision-making.

[0005] The healthcare field urgently needs a predictive method that can deeply integrate multi-source health data, quantify disease interactions, and dynamically generate personalized management strategies to address the increasingly serious public health challenge of multiple diseases among the elderly. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for predicting multiple comorbidities in the elderly, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a method for predicting multiple comorbidities in the elderly, the method comprising:

[0008] Acquire multimodal health monitoring data of the target elderly within a preset time period. The multimodal health monitoring data includes time-series data of physiological indicators, medication record sequences, and daily activity ability assessment results.

[0009] Cross-modal association analysis is performed on the multimodal health monitoring data to generate a disease interaction feature matrix, which includes the association degree of metabolic diseases, the synergistic index of circulatory system diseases, and the progression rate of neurodegenerative diseases.

[0010] A dynamic weight allocation algorithm is used to perform hierarchical clustering on the disease interaction feature matrix, and output the potential comorbidity combinations of the target elderly and the priority score of each comorbidity combination;

[0011] A set of personalized intervention strategies is generated based on the priority score. The set of personalized intervention strategies includes drug interaction avoidance strategies, rehabilitation training intensity adjustment strategies, and nutritional intake ratio strategies.

[0012] Preferably, the specific steps for performing cross-modal correlation analysis on the multimodal health monitoring data include:

[0013] Extract blood glucose fluctuation curves, blood pressure coefficient of variation, and frequency of blood oxygen saturation decrease events from the time-series data of the aforementioned physiological indicators;

[0014] The drug use record sequences are classified into sub-sequences of metabolic regulation drugs, cardiovascular drugs, and neuroprotective drugs according to their drug action mechanisms;

[0015] The correlation between the declining trend of limb coordination in the assessment results of daily activity abilities and the changes in cognitive function scores was analyzed.

[0016] The blood glucose fluctuation curve is matched with the sub-sequence of metabolic regulation drugs by time alignment to generate the first type of association feature;

[0017] The blood pressure variation coefficient and the medication time points of the cardiovascular drug subsequence are analyzed by event triggering to generate a second type of association feature;

[0018] A third type of association feature was constructed based on the decline trend in limb coordination and the dose adjustment records of neuroprotective drug sub-sequences;

[0019] The disease interaction feature matrix is ​​generated by integrating the first type of association features, the second type of association features, and the third type of association features.

[0020] Preferably, the specific steps of performing hierarchical clustering processing on the disease interaction feature matrix using a dynamic weight allocation algorithm include:

[0021] The metabolic disease weight coefficients are initialized based on the degree of matching between the correlation between the metabolic diseases and the historical comorbidity case database.

[0022] Based on the change in the circulatory system disease synergy index during the most recent monitoring period, the cardiovascular disease weight coefficient is dynamically adjusted.

[0023] The accelerating trend of the progression rate of the neurodegenerative diseases was analyzed using a sliding time window, and the weight decay factor of the neurodegenerative diseases was calculated.

[0024] The weight coefficients of metabolic diseases, cardiovascular diseases, and neurological diseases are input into a hierarchical clustering model to generate a first cluster, a second cluster, and a third cluster.

[0025] The Manhattan distance is calculated for disease combinations within each cluster, and the potential comorbidity combinations and priority scores are output.

[0026] Preferably, the specific steps for generating a personalized intervention strategy set based on the priority score include:

[0027] For the metabolic disease combinations in the first cluster, the most frequently occurring incompatible combinations are retrieved from the drug interaction database to generate the drug interaction avoidance scheme;

[0028] Based on the priority score of circulatory system diseases in the second cluster, the rehabilitation training intensity adjustment scheme corresponding to the intensity level is selected from the rehabilitation training template library;

[0029] By combining the neurodegenerative disease progression rate and nutritional intervention effect database of the third cluster, the nutritional intake ratio scheme is matched.

[0030] Preferably, the method further includes the following steps:

[0031] Real-time monitoring of physiological response data of the target elderly after implementing the personalized intervention strategy set;

[0032] When the peak-to-trough difference of the blood glucose fluctuation curve exceeds a preset threshold, the correlation of metabolic diseases is recalculated.

[0033] If the diurnal variation rate of the blood pressure variability coefficient continues to increase, then update the calculation parameters of the circulatory system disease synergy index.

[0034] The potential comorbidity combinations are dynamically corrected based on the updated disease interaction feature matrix.

[0035] Preferably, the specific steps for recalculating the association degree of metabolic diseases include:

[0036] Extract the change in insulin resistance index of the target elderly population during the current intervention period;

[0037] Obtain dose adjustment records for subsequences of metabolic regulation drugs taken concurrently;

[0038] Establish a time lag correlation model between the change in the insulin resistance index and drug dosage adjustment;

[0039] The correlation of metabolic diseases is corrected based on the output of the time lag correlation model.

[0040] Preferably, the specific steps for calculating the updated circulatory system disease synergy index include:

[0041] The duration and time of occurrence of the target elderly individuals were collected from their three most recent nocturnal blood pressure drops.

[0042] Analyze the phase difference between the nocturnal blood pressure drop event and the administration time of the sustained-release formulation in the cardiovascular drug subsequence;

[0043] The weights of the autonomic nervous system regulatory factors in the circulatory system disease synergy index are adjusted based on the phase difference.

[0044] Preferably, the method further includes the following steps:

[0045] Construct a cross-institutional knowledge graph of comorbidities in the elderly, which includes typical comorbidity evolution paths reported by different medical institutions;

[0046] The disease interaction feature matrix of the target elderly is matched with the knowledge graph using a subgraph.

[0047] When the similarity to a known high-risk comorbidity pathway is detected to exceed a threshold, an emergency consultation marker is added to the set of personalized intervention strategies.

[0048] Preferably, the specific steps of performing subgraph matching between the disease interaction feature matrix of the target elderly and the knowledge graph include:

[0049] Extract a set of reference nodes from the knowledge graph that match the age segments of the target elderly person;

[0050] Calculate the cosine similarity between each feature dimension in the disease interaction feature matrix and the feature of the reference node;

[0051] Filter the top five reference nodes by similarity to construct a local matching subgraph;

[0052] The shortest pathological development path length between nodes in the local matching subgraph is analyzed and used as the criterion for judging the similarity.

[0053] Preferably, the present invention also includes a system for predicting multiple comorbidities in the elderly, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described method for predicting multiple comorbidities in the elderly.

[0054] Compared with the prior art, the beneficial effects of the present invention are:

[0055] This invention, through the systematic integration and analysis of multimodal health data from older adults, can reveal potential comorbidity risks at an earlier stage. Traditional methods often diagnose diseases only after clinical manifestations are evident, while this invention, through comprehensive analysis of physiological indicator trends, medication patterns, and changes in activity levels, can identify abnormal patterns at an earlier stage, gaining valuable time for intervention. This prospective risk assessment helps shift the focus of health management from treatment to prevention, aligning with modern health management principles.

[0056] This method enables in-depth quantification of the interactions between different diseases. Unlike simply listing diseases, the disease interaction feature matrix generated by this invention reveals the intrinsic connections and degree of influence between metabolic, circulatory, and neurological diseases. For example, the system can quantify the potential impact of diabetes on cardiovascular health or assess the possible accelerating effect of a certain medication regimen on cognitive decline. This deep-level correlation analysis provides data support for understanding the complex mechanisms of comorbidities in the elderly.

[0057] The application of dynamic weight allocation algorithms makes the prediction results more individualized and adaptable. The health status of the elderly is a dynamically changing process. The algorithm used in this invention can adjust model parameters based on the latest characteristics of individual data, making the output comorbidity combinations and priority scores more closely reflect the individual's real-time health status. This dynamic adjustment capability avoids the limitations of a one-size-fits-all approach, allowing the prediction results to continuously optimize over time and with the accumulation of data.

[0058] In generating intervention strategies, this invention embodies comprehensiveness and personalization. The output set of intervention strategies not only focuses on medication adjustments but also encompasses multiple aspects such as rehabilitation training and nutritional balance, forming a multi-dimensional health management plan. This approach recognizes that managing multiple comorbidities requires a holistic approach, as single-faceted interventions often have limited effectiveness. Strategy generation is based on individual priority scores, ensuring a match between intervention measures and risk levels.

[0059] This method also possesses strong scalability and adaptability. Its data-driven nature allows it to continuously incorporate new analytical dimensions as new types of health data and medical discoveries emerge. The framework's flexibility allows it to adapt to different clinical scenarios and population characteristics, possessing the potential for wider application. This adaptability is of great significance in addressing ever-changing healthcare needs.

[0060] The application of this invention helps optimize the allocation of medical resources and improve the efficiency of elderly health management. By accurately identifying high-risk individuals and core intervention targets, limited medical resources can be more concentrated on the people who need them most and the most critical aspects. For elderly individuals, personalized management plans developed based on this method are expected to reduce the risks of multiple medications, delay functional decline, and thus improve their overall quality of life and health. Attached Figure Description

[0061] Figure 1 This is a schematic diagram illustrating the working principle of the method for predicting multiple comorbidities in the elderly as described in this invention.

[0062] Figure 2 A flowchart for cross-modal correlation analysis of multimodal health monitoring data;

[0063] Figure 3 A flowchart for dynamic weighted hierarchical clustering of disease interaction feature matrices;

[0064] Figure 4 A graph for evaluating the stability of the dynamic weight allocation algorithm. Detailed Implementation

[0065] 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 some embodiments of the present invention, and not all embodiments. 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.

[0066] Please see Figure 1This method integrates multi-source health data to achieve precise prediction and intervention of comorbidities in the elderly. It acquires multimodal health data from target elderly individuals over a pre-defined period, including time-series physiological indicators from wearable devices, medication records from electronic health records, and daily activity ability assessments using standardized scales. Cross-modal association analysis is then performed to generate a disease interaction feature matrix containing associations with metabolic diseases, synergistic indices for circulatory system diseases, and progression rates of neurodegenerative diseases. A dynamic weighting algorithm is used to perform hierarchical clustering on the feature matrix, dynamically adjusting feature weights based on data timeliness and clinical value. After clustering, potential comorbidity combinations and their priority scores are output. Finally, personalized intervention strategies are generated based on the scores, including drug interaction avoidance programs, individualized rehabilitation training programs, and nutritional ratio programs, forming a closed-loop management system from monitoring to intervention.

[0067] Example 1: See Figure 2 The specific process of cross-modal association analysis of multimodal health monitoring data aims to extract meaningful association features from heterogeneous data sources and construct a disease interaction feature matrix. Cross-modal association analysis begins with the in-depth processing of time-series physiological indicator data, which typically originates from continuously worn medical-grade sensors or periodically performed clinical examinations, forming long-term raw records. Extracting clinically significant feature indicators from this massive amount of time-series data is the first step. For blood glucose data, a sliding window algorithm is used to identify daily peaks and troughs, connecting these key points to form a fluctuation curve reflecting individualized blood glucose control. This curve not only shows absolute blood glucose levels but also reveals its dynamic change patterns. Blood pressure data processing focuses on its variability. The coefficient of variation is obtained by calculating the ratio of the standard deviation to the mean of blood pressure readings within a specific time period. This coefficient quantifies blood pressure stability and indirectly reflects autonomic nervous system regulation. Blood oxygen saturation data is used to statistically analyze the frequency of specific events. After setting clinically acceptable threshold ranges, the system automatically marks and counts events where blood oxygen levels fall below the threshold within a unit of time. The frequency of blood oxygen saturation decline events thus becomes an important indicator of potential abnormalities in respiratory or circulatory function.

[0068] Medication record sequence analysis requires structured classification based on pharmacological knowledge. While original records contain fields such as drug name, dosage, and administration time, they lack disease-specific information. Each record is labeled using an internationally accepted classification system, categorizing drugs into metabolic regulation, cardiovascular, and neuroprotective categories to form drug subsequences with clear therapeutic orientations. The metabolic regulation category includes records of hypoglycemic and lipid-lowering drugs; the cardiovascular category covers antihypertensive and antianginal drugs; and the neuroprotective category includes drugs that improve cognition and delay neurodegeneration. This classification directly links drug information with specific disease-specific physiological indicators. Daily activity ability assessment analysis focuses on the intrinsic relationships between functional indicators, using standardized tools such as the Barthel Index to regularly collect data and construct time series. The decline in limb coordination is quantified by analyzing the slope of changes in motor function indicators, and changes in cognitive function scores are obtained by comparing neuropsychological test results. The correlation coefficient between limb coordination and cognitive function scores at different time points is calculated to reveal whether motor function decline and cognitive impairment are synchronized, providing clues for the early identification of neurodegenerative diseases.

[0069] Association analysis between different modalities requires addressing the issue of time synchronization. The matching of blood glucose fluctuation curves with metabolic regulation drug subsequences employs a dynamic time warping algorithm. This algorithm non-linearly aligns blood glucose measurement points with discrete medication time points at different sampling frequencies. Based on this time alignment, it analyzes the morphological changes of the blood glucose curve within a specific time window after drug administration, such as calculating the rate and magnitude of blood glucose decline within one hour after rapid-acting insulin injection. This matching can generate characteristic indicators reflecting individual differences in drug efficacy, such as blood glucose response sensitivity or drug onset delay time; these indicators constitute the first type of association feature. Association analysis of blood pressure variability coefficient (CVC) with cardiovascular drug subsequences adopts an event-triggered approach, setting an observation window with each medication time point as the event starting point. It monitors whether there are statistically significant changes in CVC within a specific time period after medication, such as the increase or decrease trend of CVC within four hours after taking angiotensin-converting enzyme inhibitors. By comparing the difference in CVC before and after medication, a blood pressure stability response index is calculated. This index characterizes the immediate effect of the drug on blood pressure stability and forms the second type of association feature. For sustained-release formulations, the observation window needs to be extended to capture their long-term effects.

[0070] The correlation analysis of neurological function-related data focuses on the coordination of long-term trends, with a time-series correlation between the decline in limb coordination and records of neuroprotective drug dosage adjustments. A regression model is established to correlate the slope of limb coordination changes over a subsequent period with the time of dosage adjustment, assessing whether an increase in drug dosage is accompanied by a slowdown in the rate of coordination decline. For example, analyzing the rate of change in balance test scores over three months after increasing donepezil dosage reveals the modifying effect of the drug on functional decline and generates a third type of correlation feature. Dosage adjustments include not only increases but also decreases or discontinuation to comprehensively assess the impact. The final feature matrix integration needs to address the differences in the dimensions and value ranges of different features, and the three types of correlation features are standardized to ensure comparability. The blood glucose response sensitivity (first type of correlation feature), the blood pressure stability response index (second type of correlation feature), and the functional decline modification coefficient (third type of correlation feature) are mapped to a unified numerical range. The correlation between metabolic diseases is primarily derived from glycemic response sensitivity and other synergistic metabolic indicators. The comorbidity index for circulatory system diseases is calculated based on parameters such as blood pressure stability response index and heart rate variability. The progression rate of neurodegenerative diseases integrates functional decline modification coefficients and cognitive score change rates. These feature values ​​are organized into a matrix according to time points, with each row representing a monitoring time point and each column corresponding to a feature dimension, thus forming a comprehensive panoramic view that reflects the dynamic interactions between diseases. This feature matrix serves as the foundational data structure for subsequent cluster analysis, and its quality directly affects the accuracy of comorbidity pattern identification.

[0071] Example 2: See Figure 3The application of the dynamic weight allocation algorithm begins with the analysis of the association between metabolic diseases. This association data originates from the feature matrix generated during the cross-modal association analysis phase. When initializing the metabolic disease weight coefficients, the system accesses a historical comorbidity case database, which stores a large amount of clinically confirmed comorbidity case data from elderly individuals and their evolution trajectories. The system calculates the matching degree between the current target elderly individual's metabolic disease association value and the typical association values ​​of the same age group in the case database. The matching degree is quantified by measuring the deviation of the current value from the historical data distribution range. Specifically, a statistical comparison method is used: the system calculates the absolute deviation of the current association value from the median of the historical distribution and assesses its relative position to the historical data range. If the current value falls within the common interval of the historical distribution, the matching degree is high, indicating that the individual's risk pattern is similar to the typical pattern of the group; if the current value deviates significantly from the historical range, the matching degree is low, suggesting an abnormal risk signal. This quantification process allows the system to initialize the metabolic disease weight coefficients based on the matching degree between individual and group patterns, thereby enhancing the individualized adaptability of the cluster analysis. If the current correlation is within the historically common range, a lower initial weight coefficient is assigned; if it exhibits rare high correlation characteristics, its weight coefficient is increased accordingly, enabling the algorithm to focus on abnormal signals. The historical case database is continuously updated to ensure that the weight initialization is based on the latest clinical practice knowledge.

[0072] The dynamic adjustment of the Circulatory System Disease Synergistic Index relies on identifying its temporal variation patterns. The system extracts the index value sequence from the most recent complete monitoring period. Calculating the magnitude of change in this sequence involves comparing the numerical differences between the start and end points of the period, and considering the fluctuations throughout the period. Specifically, when calculating the magnitude of change in the Circulatory System Disease Synergistic Index sequence, the system selects the index value sequence from the most recent complete monitoring period, for example, a seven-day period. The magnitude of change is calculated by comparing the numerical differences between the start and end points of the period to obtain the absolute change, which is the difference between the end point value and the start point value. The system considers the fluctuations throughout the period, assessing the stability of the changes by analyzing the dispersion of the data points around the trend line or calculating the coefficient of variation. For example, the system uses a sliding window technique to identify peaks and troughs in the sequence and statistically analyzes the frequency and amplitude of fluctuations to capture intermediate fluctuation patterns. This comprehensive approach avoids overlooking instantaneous changes that may be missed by relying solely on endpoint differences, thus more comprehensively reflecting the dynamic evolution of the index. The calculated magnitude of change is used to dynamically adjust the cardiovascular disease weight coefficient. Large changes indicate an unstable circulatory system, and the weight is increased accordingly to prioritize this area. Slow changes maintain or slightly decrease the weight. This ensures the model's real-time responsiveness to individual health conditions. If a rapid increase or decrease in the synergy index is detected, indicating a significant change in the circulatory system, the cardiovascular disease weight coefficient is dynamically increased to make the cluster analysis focus more on such acute changes. If the index remains stable or fluctuates only within a small range, the current weight is maintained or slightly decreased to avoid overreacting to static signals. This dynamic adjustment mechanism allows the model to adapt to the real-time evolution of the patient's condition and capture instantaneous increases in cardiovascular risk. The progression rate of neurodegenerative diseases is analyzed using a sliding time window technique, with the window size set at seven days based on clinical experience. The first and second derivatives of the progression rate are calculated within each sliding window; the first derivative reflects the rate of change, and the second derivative reveals the acceleration of change. When a continuously accelerating progression rate is detected, it indicates that the neurodegenerative process may be entering an active phase, at which point a neurodegenerative disease weight decay factor is applied. The decay factor is calculated based on the duration and magnitude of acceleration. The more pronounced and longer the acceleration, the smaller the decay factor, thus reducing the influence of this dimension in clustering. This design enables the model to distinguish between the chronic stable phase and the acute progressive phase of the disease, and to allocate analytical weights appropriately.

[0073] After weighting, the weighted feature matrix is ​​input into a hierarchical clustering model for processing. The clustering process employs a bottom-up aggregation strategy, initially treating each data point as an independent cluster and iteratively merging the two most similar clusters. Similarity is measured using a weighted Euclidean distance formula, where the weights of each feature dimension are determined by the aforementioned dynamic weighting algorithm. The clustering termination condition is jointly determined by the inter-cluster distance threshold and clinical interpretability, typically generating three main clusters corresponding to different comorbidity patterns. The first cluster primarily contains comorbidity features dominated by metabolic abnormalities, the second cluster is characterized by synergistic changes in circulatory system diseases, and the third cluster focuses on neurodegenerative-related problem combinations. Analysis of the internal structure of each cluster is achieved through Manhattan distance calculation, a distance metric that effectively captures grid-like distribution patterns in high-dimensional space. The Manhattan distance between each data point within a cluster and the cluster center is calculated; the distribution of this distance reflects the tightness and consistency of the comorbidity pattern. Based on the distance distribution statistics, a priority score is generated for each potential comorbidity combination; data points closer to the cluster center are considered to represent more typical comorbidity patterns and are assigned higher priority accordingly. Clinical risk factors, such as the urgency or prognostic impact of certain disease combinations, should also be considered when appropriately adjusting the scores.

[0074] The generation of personalized intervention strategy sets begins with in-depth analysis of the first cluster, which is primarily characterized by combinations of metabolic diseases. Developing drug interaction avoidance strategies requires querying multi-source drug knowledge bases, including pharmacology databases and clinical guideline databases. The system identifies potential contraindications or adverse interactions in the current medication regimen, particularly considering special considerations for elderly patients with impaired liver or kidney function. Avoidance strategies not only provide alternative medication recommendations but also detail the timing of medication changes and dosage adjustments to ensure a safe transition. For complex medication situations, the system marks steps requiring pharmacist review to ensure the feasibility of the recommendations.

[0075] The second cluster, representing comorbidity patterns of circulatory system diseases, requires personalized rehabilitation training programs. The intensity of rehabilitation training is adjusted based on priority scores and the patient's individual functional status. The system's built-in rehabilitation training template library contains exercise prescriptions at different intensity levels, ranging from low-intensity bed activities to moderate-to-high-intensity aerobic training. Patients with higher priority scores may receive more aggressive exercise recommendations, but will strictly adhere to the principle of gradual progression. Each training program clearly specifies frequency, intensity, and time parameters, along with necessary monitoring indicators and safety precautions. The programs also consider the patient's activity preferences and environmental conditions to improve adherence. The third cluster, representing comorbidity patterns related to neurodegenerative diseases, requires targeted nutritional interventions. The nutritional intake ratios are developed with reference to evidence-based medicine. The system analyzes the correlation between the current disease progression rate and the intake of various nutrients, generating personalized recommendations based on evidence from the nutritional intervention effectiveness database. The ratios not only focus on macronutrient balance but also provide detailed recommendations for the intake of specific fatty acids, vitamins, and minerals. Considering common problems such as dysphagia and cognitive impairment, the programs offer practical suggestions for adjusting food texture and optimizing the dining environment. All nutritional recommendations take into account the patient's dietary habits and cultural background to ensure feasibility.

[0076] See Figure 4In the prediction of comorbidities in the elderly, the dynamic weight allocation algorithm involves dynamic adjustment and stability assessment mechanisms for weight coefficients. Specifically, the dynamic weight allocation trend chart displays the changes in weights for metabolic diseases, cardiovascular diseases, and neurological diseases within a preset monitoring period (weeks 1-7), reflecting the real-time optimization process of the algorithm based on multimodal health monitoring data for the weight coefficients of various diseases. The weight adjustment stability assessment chart quantifies the volatility of the adjustment process by calculating the year-on-year change rate of weights in adjacent periods, using a zero baseline as a reference; the closer the change rate is to zero, the more robust the weight allocation. In the algorithm implementation, weight initialization relies on the matching degree analysis of the historical comorbidity case database. The matching degree is calculated using the standardized deviation of the current disease correlation degree and the median value of the historical data distribution. For the cardiovascular disease weight, dynamic adjustment is based on the change amplitude of the circulatory system disease synergy index within the monitoring period, and the change amplitude is determined through a comprehensive assessment of sequence endpoint differences and internal volatility. The decay factor for the neurological disease weight uses a sliding time window technique to analyze the acceleration trend of the progression rate, with the window size set to 7 days to match the clinical monitoring period. The weighted feature matrix is ​​input into a hierarchical clustering model, and similarity is measured using dynamically weighted Euclidean distance. Clusters representing different comorbidity patterns are iteratively generated. During parameter configuration, the monitoring period is defined as 7 days, and the adjustment threshold for the weight coefficients is set to the statistical significance level of the rate of change; for example, a recalculation is initiated when the month-on-month change rate consistently exceeds the historical fluctuation range. Intra-cluster consistency in cluster analysis is evaluated using Manhattan distance, which is calculated across all feature dimensions to optimize the priority scoring of comorbidity combinations.

[0077] Example 3: The system starts after the personalized intervention strategy is implemented. It continuously collects and analyzes physiological indicator response data of the target elderly individuals to achieve iterative optimization of the predictive model. Real-time monitoring relies on a network of IoT devices deployed in the elderly's living environment, including wireless blood glucose meters, smart blood pressure monitors, and pulse oximeters. These devices collect data at preset frequencies and send it to the central processing unit via encrypted transmission protocols. After data inflow, it undergoes quality control and outlier filtering to ensure the reliability of subsequent analysis. The monitoring cycle is typically set to 24 hours to match daily physiological rhythms. When the system detects that the peak-to-trough difference in the blood glucose fluctuation curve exceeds a preset threshold, it automatically triggers a recalculation process for the correlation of metabolic diseases. The preset threshold is dynamically adjusted based on clinical guidelines; for example, for diabetic patients, the daily blood glucose difference threshold is set to 50 mg / dL. Exceeding this threshold indicates a potential deterioration in blood glucose control. The recalculation process extracts the change in the insulin resistance index within the current intervention cycle. The insulin resistance index is calculated using a steady-state model assessment method, using fasting blood glucose and insulin concentration as input parameters. The change is expressed as the difference between the index at the end of the current cycle and the beginning of the cycle. Simultaneously, subsequences of metabolic regulatory drugs taken during the same period were obtained from electronic health records, including drug names, dosage adjustment times and magnitudes, to establish a time-lag correlation model between changes in the insulin resistance index and drug dosage adjustments. This model employs distributed lag regression, setting a maximum lag time of 7 days to capture the delayed effects of drug influence. The model outputs a correlation coefficient to quantify the degree of influence of dosage adjustments on the resistance index. The correlation coefficient's magnitude and direction are used to adjust the association degree for metabolic diseases; for example, when the coefficient shows a strong positive correlation, the correlation weight is appropriately increased to reflect the importance of drug intervention.

[0078] Monitoring the blood pressure variability coefficient focuses on its diurnal rhythm variation. The system calculates the diurnal difference rate by comparing the nighttime and daytime blood pressure variability coefficients, which is used to assess the stability of autonomic nervous system function. Specifically, the system defines daytime and nighttime periods, such as 06:00 to 22:00 during the day and 22:00 to 06:00 the next day at night, and adjusts these periods individually according to individual sleep schedules. The nighttime and daytime blood pressure variability coefficients are calculated separately, with the coefficient of variation obtained by dividing the standard deviation of systolic blood pressure readings within that period by the mean. The diurnal difference rate is obtained by comparing the nighttime and daytime variability coefficients, reflecting the stability of the blood pressure rhythm: a ratio close to 1 indicates a stable diurnal rhythm; a ratio significantly deviating from 1 suggests potential autonomic nervous system dysfunction. The system continuously monitors this difference rate and assesses its trend. For example, a continuously increasing difference rate indicates deteriorating blood pressure stability, triggering parameter updates in the circulatory system disease synergy index. This monitoring mechanism helps identify cardiovascular risks early and provides data support for dynamically revising comorbidity predictions. A diurnal variation rate increasing by more than 5% for three consecutive days is considered a signal of continued deterioration, triggering an update of the calculation parameters for the circulatory system disease synergy index. The parameter update process begins by collecting data on the three most recent nocturnal blood pressure drops, defined as a decrease in systolic blood pressure exceeding 20 mmHg and lasting for more than 10 minutes. The start time, duration, and magnitude of the blood pressure drop for each event are recorded. The phase difference between these events and the administration time of sustained-release formulations in the cardiovascular drug subsequence is analyzed. The phase difference is calculated as the absolute difference between the event occurrence time and the most recent medication time, expressed in hours. The weights of autonomic nervous system regulatory factors in the synergy index are adjusted based on the phase difference. For example, when the phase difference is less than 2 hours, it indicates a possible correlation between medication time and blood pressure drops, and the weight is reduced accordingly to weaken its impact. The entire parameter update process employs a sliding window analysis with a window size set to 72 hours to ensure data timeliness.

[0079] The core of the dynamic correction mechanism lies in reassessing potential comorbidity combinations based on the updated disease interaction feature matrix. The feature matrix is ​​updated every 24 hours to balance computational efficiency and model sensitivity. The correction process re-runs the hierarchical clustering algorithm, using the latest weight assignments and eigenvalues. The clustering results are compared with previous versions to identify newly added or disappeared comorbidity patterns. For combinations with significantly changed priority scores, the system automatically adjusts the priority of their intervention strategies, for example, elevating newly emerging high-risk combinations to the top of the treatment queue. The corrected comorbidity list is presented to healthcare professionals through a visual interface, highlighting the changes for quick identification of key issues.

[0080] The integration of real-time monitoring data streams and the early warning module adopts an event-driven architecture. When an abnormal pattern is detected in the data stream, the corresponding calculation process is immediately triggered, reducing the delay of manual intervention. The early warning module is configured with a multi-level threshold system, including warning and critical levels. Different levels trigger different response mechanisms; the warning level only logs the data, while the critical level sends an immediate alert to medical staff. The data persistence layer stores all historical monitoring records and intermediate calculation results, supporting retrospective analysis and trend prediction. The storage format uses a time-series database to optimize query performance. In the blood glucose monitoring stage, in addition to focusing on the peak-to-trough difference, the system also analyzes the morphological characteristics of the blood glucose fluctuation curve, such as the area under the curve, fluctuation frequency, and amplitude variation coefficient. These additional indicators provide a more comprehensive assessment of blood glucose control; for example, the area under the curve reflects the overall glycemic load, and the fluctuation frequency indicates blood glucose stability. When multiple indicators are abnormal simultaneously, the system increases the trigger sensitivity and initiates a recalculation process in advance to prevent risks. The calibration of blood glucose data takes into account equipment errors and individual differences, and is regularly corrected by comparing with venous blood test results.

[0081] In terms of blood pressure monitoring, the calculation of diurnal variation rate strictly distinguishes between daytime and nighttime periods. Daytime is defined as 06:00 to 22:00, and nighttime is 22:00 to 06:00 the next day, with blood pressure variation coefficients calculated segmentally. The system also incorporates postural change impact analysis, recording blood pressure response data from supine to standing positions to identify potential problems such as orthostatic hypotension. For patients taking multiple cardiovascular medications, phase difference analysis considers the potential synergistic effects of drug interactions; for example, the combined use of two antihypertensive drugs may exacerbate nighttime blood pressure drops. Clustering algorithm optimization during dynamic correction includes adjusting distance metric parameters and heuristic rules for cluster number to adapt to changes in data distribution. The algorithm incorporates a forgetting factor mechanism, gradually reducing the weight of historical data to make the model more focused on recent trends. The reliability of the corrected comorbidity combinations is assessed using confidence scores; low-confidence combinations require manual review and confirmation before being included in the intervention strategy.

[0082] The system adopts a microservice architecture, with monitoring, analysis, and correction modules deployed independently. Asynchronous communication via message queues enhances system resilience and scalability. Each module has self-monitoring capabilities, automatically switching to a backup instance in case of anomalies to ensure service continuity. Data security is implemented with end-to-end encryption and access control, complying with medical data protection regulations. The closed-loop design of the real-time monitoring and dynamic correction mechanism allows the predictive model to continuously learn individual health change patterns, gradually improving the accuracy of comorbidity prediction. Through continuous iterative optimization, the system provides more targeted health management support for the elderly, adapting to the needs of long-term disease management. (Diurnal variation rate) The calculation formula is:

[0083] ;

[0084] in: The coefficient of variation for nighttime blood pressure is obtained by dividing the standard deviation of systolic blood pressure readings during the nighttime period by the mean. This represents the coefficient of variation in daytime blood pressure, and the calculation method is similar. It represents the diurnal variation rate, reflecting the stability of the diurnal rhythm of blood pressure.

[0085] Implementation details include filtering and denoising of the raw blood pressure signal during data preprocessing, using a moving average filter to smooth random fluctuations and preserve physiological trends. Before calculating the coefficient of variation, significant artifacts such as motion artifacts or data loss due to equipment malfunction are excluded to ensure the coefficient reflects the true physiological state. Diurnal segmentation is personalized based on individual sleep schedules; for night shift workers or those with sleep disorders, activity diaries are used to assist segmentation, improving accuracy. Recalculation and parameter update triggering logic is embedded in a complex event processing engine, supporting multi-condition combinations, such as triggering the process only when both a blood glucose threshold breakthrough and a specific medication usage record are met. The engine uses a rule base to define triggering conditions, which can be updated online to adapt to new clinical evidence. Processing results are written to a distributed database, supporting high-concurrency access and fast retrieval.

[0086] The system integrates with existing electronic health record systems, automatically acquiring medication records and diagnostic information to reduce duplicate data entry errors. Integration utilizes standard medical data exchange protocols to ensure interoperability. The user interface design considers the usage habits of elderly users, providing large fonts and voice prompts to enhance usability. The calibration and maintenance of monitoring devices are incorporated into the system management, with regular reminders for sensor or battery replacements to ensure data quality. Device data is transmitted via Bluetooth or Wi-Fi, cached locally when offline, and automatically synchronized upon network recovery to prevent data loss. The dynamic correction mechanism's evaluation module regularly generates model performance reports, comparing the consistency of predicted results with actual health events, and using metrics such as sensitivity and specificity to quantify model effectiveness. These reports help healthcare professionals understand system limitations and use predicted results appropriately. Emphasis is placed on user experience and clinical integration, with continuous improvement to make the system a reliable tool for elderly health management. The technical solution balances advanced technology with practicality, ensuring stable operation in real-world scenarios.

[0087] Example 4: Referring to Table 1, the complete process of recalculating the metabolic disease association and updating the circulatory system disease synergistic index parameters is explained in detail through the following specific case. Assume a 72-year-old male patient, Mr. Zhang, with a history of type 2 diabetes and hypertension, currently receiving metformin and enalapril treatment. The system triggered the relevant recalculation mechanism during the past week of monitoring. The recalculation of the metabolic disease association begins with the extraction of changes in the insulin resistance index during the current intervention period, which is set at 7 days, from March 1st to March 7th, 2024. Fasting venous blood is collected every morning to measure blood glucose and insulin levels, and the daily insulin resistance index is calculated using a steady-state model assessment method. Table 1 shows the changes in key indicators during this period, including date, fasting blood glucose, fasting insulin, calculated insulin resistance index, and daily metformin dose record.

[0088] Table 1: Metabolic Indicators and Medication Records During the Intervention Period

[0089] As shown in Table 1, the insulin resistance index (IRI) increased significantly on March 5th, the same day the metformin dose was increased from 1000mg to 1500mg. The change in IRI was calculated as the difference between the value at the end of the cycle (March 7th) and the value at the beginning of the cycle (March 1st). In this example, the change was 0.07, indicating a slight increase in insulin resistance. When establishing a time-lag correlation model between the change in IRI and drug dosage adjustment, the system used the dosage adjustment date (March 5th) as the zero point and analyzed the index change pattern for three days before and after the adjustment. The model detected that within two days after the dosage increase, the IRI first increased and then decreased, exhibiting a certain delayed response pattern. This time-lag correlation was quantified by calculating the difference in the slope of the index change before and after the dosage adjustment. The results showed that the rate of index change temporarily increased after the dosage increase, indicating that a metabolic adaptation response may occur in the early stages of drug adjustment.

[0090] Based on the output of the time-lag correlation model, the system corrected the correlation for metabolic diseases. The original correlation value was 0.65 (range 0-1, with higher values ​​indicating stronger correlations). Based on the model's observation of increased insulin resistance short-term after dose adjustment, the correlation was corrected to 0.72. The corrected correlation value more accurately reflects the instability of the metabolic state under current drug intervention, suggesting the need for enhanced blood glucose monitoring and possible adjustments to the treatment plan. The update process for the circulatory system disease synergistic index parameters began with the monitoring of nocturnal blood pressure drops. Zhang's 24-hour ambulatory blood pressure recorder detected three significant nocturnal blood pressure drops between March 5th and March 7th. The first event occurred at 23:15 on March 5th, when systolic blood pressure dropped from 145 mmHg to 118 mmHg within 15 minutes, remaining stable for 25 minutes before recovering. The second event occurred at 22:50 on March 6th, when systolic blood pressure dropped from 138 mmHg to 115 mmHg, remaining stable for 30 minutes. The third event occurred at 23:30 on March 7th, when systolic blood pressure dropped from 142 mmHg to 119 mmHg, remaining stable for 22 minutes. All of these events met the preset criteria for a sudden drop in nocturnal blood pressure (a decrease in systolic blood pressure of more than 20 mmHg, lasting for more than 10 minutes).

[0091] This study analyzed the phase difference between nocturnal blood pressure drops and the timing of cardiovascular medication administration. Mr. Zhang was taking enalapril, a sustained-release formulation, daily at 9:00 PM. The phase differences between the time of each event and the medication administration time were calculated, yielding phase differences of 2.25 hours, 1.83 hours, and 2.5 hours for the three events. Phase difference analysis showed that the blood pressure drops were concentrated approximately 2 hours after medication administration, indicating a potential temporal correlation between peak drug concentration and blood pressure fluctuations. The weight of the autonomic nervous system regulation factor in the circulatory system disease synergy index was adjusted based on the phase difference; the original weight was 0.6. The average phase difference of the three events was 2.19 hours, close to the 2-hour threshold, so the system lowered the weight of the autonomic nervous system regulation factor to 0.55. This weight adjustment reflects the overlap between the peak drug effect period and the period of blood pressure instability, suggesting a need to reassess the medication administration time or consider adjusting the drug formulation.

[0092] The entire recalculation and parameter update process is completed automatically in the system backend. Medical staff can view the calculation logs and adjustment basis on the front-end interface. The interface displays the sequence of key events in a timeline format, including the time point of dose adjustment, the time of abnormal physiological indicators, the calculation trigger time, and the parameter update results. Medical staff can trace the data support for each decision node and understand the logical basis of the system's judgment. In the case of Mr. Zhang, the recalculation results of the metabolic disease correlation showed that although the metformin dose was increased, the improvement in insulin resistance was not significant, suggesting that there may be decreased drug sensitivity or other influencing factors. The system suggested that clinicians pay attention to possible factors that aggravate insulin resistance, such as weight changes, dietary adjustments, or the effects of accompanying medications. After the circulatory system disease synergistic index parameters are updated, the system generates nighttime blood pressure management suggestions, including adjusting the medication time to an earlier time (such as 8:00 pm) or considering splitting the medication to smooth out fluctuations in blood drug concentration.

[0093] Data quality control is particularly crucial. The system flags and verifies abnormal measurements. For example, in blood pressure monitoring, a single drastic change needs to exclude measurement errors or postural disturbances; only a continuous and stable downward trend is considered a valid event. Blood glucose and insulin testing require the use of the same methods and reagents in the same laboratory to avoid errors introduced by test variability. The accuracy of medication dosage recording is ensured through multi-source validation. The system not only collects electronic prescription information but also records actual medication administration via smart pillboxes and cross-checks it with patient self-reports. Time recording accuracy is uniformly maintained in a 24-hour format, accurate to the minute. All device times are synchronized via a network time protocol to ensure the accuracy of time-series analysis. The trigger conditions for recalculation mechanisms are set with clinical practicality in mind, avoiding excessively frequent calculations that could cause interference. The system is configured with a minimum time interval protection, allowing a maximum of one recalculation of the same type within 24 hours, unless a critical value change is detected. The clinical application of calculation results requires medical staff to combine them with clinical judgment. The suggestions generated by the system are marked as reference opinions, and the final decision rests with the attending physician. The implementation method demonstrated in the case study reflects the system's adaptability to personalized treatment, dynamically adjusting assessment parameters based on the patient's response to treatment. This dynamic adjustment mechanism allows the comorbidity prediction model to maintain accuracy as the condition changes, providing continuous support for long-term disease management. The entire process design emphasizes clinical operability and safety, ensuring the effective integration of the technical solution into real-world medical environments.

[0094] Example 5: Assume a 78-year-old female patient, Ms. Wang, suffering from hypertension, type 2 diabetes, and mild cognitive impairment. Her disease interaction feature matrix shows a metabolic disease association degree of 0.71, a circulatory system disease synergy index of 0.63, and a neurodegenerative disease progression rate of 0.45. The data used to construct the cross-institutional knowledge graph of comorbidities in the elderly comes from the electronic medical record systems of multiple tertiary hospitals and community health centers. This data was anonymized before being integrated into a unified platform. The knowledge graph is stored using a graph database. Nodes represent different disease states or health conditions, and edges represent the transformation relationships or co-occurrence probabilities between diseases. Each node contains rich attribute information, such as disease name, age of onset, duration of disease, and commonly used treatment plans; edges have weight attributes, representing the conditional probability or time interval of disease development. The graph is updated regularly to incorporate the latest clinical research evidence and real-world data, maintaining its timeliness and clinical relevance.

[0095] When matching Wang's disease interaction feature matrix with the knowledge graph, the system extracts a set of reference nodes matching his age segment from the graph. The age segment is set to 75-80 years old, which contains 327 reference nodes, covering common chronic diseases in the elderly and their combinations. The matching process calculates the similarity between each dimension of the feature matrix and the features of the reference nodes, using an improved cosine similarity algorithm that considers both the absolute difference in values ​​and the consistency of trends. The similarity calculation results show that the five reference nodes closest to Wang's features correspond to common comorbidity patterns such as "hypertension-diabetes-cognitive impairment", "hypertension-coronary heart disease-mild cognitive impairment", and "diabetes-cerebrovascular disease-cognitive decline". After selecting the top five reference nodes with the highest similarity, the system constructs a local matching subgraph around them. This subgraph not only contains these five nodes but also extends to include their directly connected first-order neighbors, forming a small comorbidity evolution network. The shortest path length between nodes in the local matching subgraph is analyzed. The path length is calculated using a traversal algorithm, considering the weight and directionality of the edges. Specifically, when analyzing the shortest pathological development path length between nodes in the local matching subgraph, the system employs a graph traversal algorithm, considering edge weights and directionality. The local matching subgraph consists of the top five reference nodes with the highest similarity selected from the knowledge graph and their first-order neighbor nodes. Nodes represent disease states, and edges represent transformation relationships or co-occurrence probabilities between diseases, with weight attributes such as conditional probability or time intervals. Path length calculation uses an improved Dijkstra algorithm, traversing all possible paths from the starting node to the target node and accumulating edge weights as path costs. The algorithm prioritizes edges with smaller weights, representing disease development paths with high probability or short time intervals. Directionality ensures that the path conforms to the temporal logic of disease evolution, such as a unidirectional relationship from hypertension to diabetes. The calculated shortest path length is used as a similarity criterion; the shorter the length, the closer the comorbidity pattern of the target elderly person is to known high-risk paths, thus triggering an emergency consultation marker. This method enhances the accuracy and clinical operability of comorbidity risk prediction. In Wang's case, the path length from the "hypertension-diabetes-cognitive impairment" node to the "hypertension-diabetes-vascular dementia" node is 2, indicating the existence of an intermediate transitional state; the path length to the "hypertension-kidney damage-cognitive impairment" node is 3, indicating that the deterioration of kidney function may be a potential direction of development.

[0096] When the similarity to a known high-risk comorbidity pathway exceeds a threshold, the system adds an emergency consultation marker to the personalized intervention strategy set. The threshold is set at 0.75, determined based on a retrospective analysis of previous cases. In Wang's matching results, the similarity with the "hypertension-diabetes-cognitive impairment" node reached 0.81, exceeding the threshold, and the path length from this node to the "stroke-severe cognitive impairment" node was short, only 2. This pattern indicates that the patient is at high risk of neurovascular events, and the system automatically generates an emergency consultation marker, recommending a joint assessment by neurologists, endocrinologists, and geriatricians. The knowledge graph construction process emphasizes data quality control and standardized processing. Data from different institutions uses a unified medical terminology coding system, such as ICD-10 disease classification codes and ATC drug codes. The data integration phase addresses terminology differences and coding inconsistencies, extracting key information from unstructured text using natural language processing technology. Graph maintenance includes regularly verifying the clinical rationality of relationships between nodes, and having a team of medical experts review automatically discovered new associations to ensure the accuracy of knowledge.

[0097] The subgraph matching algorithm has been optimized for large-scale data processing, and a distributed computing framework has been adopted to improve matching efficiency. Visualizing the matching results helps healthcare professionals intuitively understand patients' comorbidity risk trajectories; the graphical interface uses different colors to indicate risk levels, with red representing high-risk paths requiring immediate attention. The system also provides explanations of the matching criteria, displaying the original case characteristics and clinical outcomes of the reference nodes to help healthcare professionals understand the system's recommendations. In a real-world application scenario, the case of Mr. Wang demonstrates how knowledge graphs can identify potential risks. The high-risk paths matched showed that a high proportion of elderly patients with similar conditions experienced cerebrovascular events or significant cognitive decline within a year. Based on this finding, the system not only marked consultation needs but also adjusted the monitoring protocol, adding cerebral hemodynamic assessments and more frequent cognitive function screenings. Drug interaction testing pays special attention to the synergistic effects of antihypertensive and hypoglycemic drugs on cerebrovascular health, avoiding interference between treatment measures.

[0098] The knowledge graph's update mechanism ensures its continuous learning capability. Newly diagnosed case data is incorporated into the graph after a certain period of verification, enriching the relationships between nodes. Wang's long-term follow-up data is also fed back to the system. If his condition progresses along the predicted path, the weight of the corresponding edge is strengthened; if a deviation occurs, the graph structure is adjusted to improve the accuracy of future predictions. Cross-institutional data sharing faces challenges in privacy protection and security management. Federated learning technology is used in implementation, with each medical institution extracting features locally and sharing only the aggregated knowledge features, not the original case data. Data transmission and storage are encrypted throughout, with strictly controlled access permissions, complying with medical data security standards. Integration of the knowledge graph with clinical decision support systems requires consideration of workflow adaptability. Prediction results are pushed to the hospital's electronic medical record system through a standard interface, embedding into existing diagnostic and treatment processes. Medical staff can conveniently view system prompts in their daily work while retaining final decision-making authority; the technical system serves as a supplementary reference. Wang's case illustrates that subgraph matching of the knowledge graph can uncover long-term risk patterns that are difficult to intuitively judge from current data alone.

[0099] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0100] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting multiple comorbidities in the elderly, characterized in that, Includes the following steps: Acquire multimodal health monitoring data of the target elderly within a preset time period. The multimodal health monitoring data includes time-series data of physiological indicators, medication record sequences, and daily activity ability assessment results. Cross-modal association analysis is performed on the multimodal health monitoring data to generate a disease interaction feature matrix, which includes the association degree of metabolic diseases, the synergistic index of circulatory system diseases, and the progression rate of neurodegenerative diseases. A dynamic weight allocation algorithm is used to perform hierarchical clustering on the disease interaction feature matrix, and output the potential comorbidity combinations of the target elderly and the priority score of each comorbidity combination; A set of personalized intervention strategies is generated based on the priority score. The set of personalized intervention strategies includes drug interaction avoidance strategies, rehabilitation training intensity adjustment strategies, and nutritional intake ratio strategies.

2. The method for predicting multiple comorbidities in the elderly according to claim 1, characterized in that, The specific steps for performing cross-modal correlation analysis on the multimodal health monitoring data include: Extract blood glucose fluctuation curves, blood pressure coefficient of variation, and frequency of blood oxygen saturation decrease events from the time-series data of the aforementioned physiological indicators; The drug use record sequences are classified into sub-sequences of metabolic regulation drugs, cardiovascular drugs, and neuroprotective drugs according to their drug action mechanisms; The correlation between the declining trend of limb coordination in the assessment results of daily activity abilities and the changes in cognitive function scores was analyzed. The blood glucose fluctuation curve is matched with the sub-sequence of metabolic regulation drugs by time alignment to generate the first type of association feature; The blood pressure variation coefficient and the medication time points of the cardiovascular drug subsequence are analyzed by event triggering to generate a second type of association feature; A third type of association feature was constructed based on the decline trend in limb coordination and the dose adjustment records of neuroprotective drug sub-sequences; The disease interaction feature matrix is ​​generated by integrating the first type of association features, the second type of association features, and the third type of association features.

3. The method for predicting multiple comorbidities in the elderly according to claim 2, characterized in that, The specific steps for performing hierarchical clustering of the disease interaction feature matrix using a dynamic weight allocation algorithm include: The metabolic disease weight coefficients are initialized based on the degree of matching between the correlation between the metabolic diseases and the historical comorbidity case database. Based on the change in the circulatory system disease synergy index during the most recent monitoring period, the cardiovascular disease weight coefficient is dynamically adjusted. The accelerating trend of the progression rate of the neurodegenerative diseases was analyzed using a sliding time window, and the weight decay factor of the neurodegenerative diseases was calculated. The weight coefficients of metabolic diseases, cardiovascular diseases, and neurological diseases are input into a hierarchical clustering model to generate a first cluster, a second cluster, and a third cluster. The Manhattan distance is calculated for disease combinations within each cluster, and the potential comorbidity combinations and priority scores are output.

4. The method for predicting multiple comorbidities in the elderly according to claim 3, characterized in that, The specific steps for generating a personalized intervention strategy set based on the priority score include: For the metabolic disease combinations in the first cluster, the most frequently occurring incompatible combinations are retrieved from the drug interaction database to generate the drug interaction avoidance scheme; Based on the priority score of circulatory system diseases in the second cluster, the rehabilitation training intensity adjustment scheme corresponding to the intensity level is selected from the rehabilitation training template library; By combining the neurodegenerative disease progression rate and nutritional intervention effect database of the third cluster, the nutritional intake ratio scheme is matched.

5. The method for predicting multiple comorbidities in the elderly according to claim 4, characterized in that, It also includes the following steps: Real-time monitoring of physiological response data of the target elderly after implementing the personalized intervention strategy set; When the peak-to-trough difference of the blood glucose fluctuation curve exceeds a preset threshold, the correlation of metabolic diseases is recalculated. If the diurnal variation rate of the blood pressure variability coefficient continues to increase, then update the calculation parameters of the circulatory system disease synergy index. The potential comorbidity combinations are dynamically corrected based on the updated disease interaction feature matrix.

6. The method for predicting multiple comorbidities in the elderly according to claim 5, characterized in that, The specific steps for recalculating the association degree of metabolic diseases include: Extract the change in insulin resistance index of the target elderly population during the current intervention period; Obtain dose adjustment records for subsequences of metabolic regulation drugs taken concurrently; Establish a time lag correlation model between the change in the insulin resistance index and drug dosage adjustment; The correlation of metabolic diseases is corrected based on the output of the time lag correlation model.

7. The method for predicting multiple comorbidities in the elderly according to claim 6, characterized in that, The specific steps for calculating the updated circulatory system disease synergy index include: The duration and time of occurrence of the target elderly individuals were collected from their three most recent nocturnal blood pressure drops. Analyze the phase difference between the nocturnal blood pressure drop event and the administration time of the sustained-release formulation in the cardiovascular drug subsequence; The weights of the autonomic nervous system regulatory factors in the circulatory system disease synergy index are adjusted based on the phase difference.

8. The method for predicting multiple comorbidities in the elderly according to claim 7, characterized in that, It also includes the following steps: Construct a cross-institutional knowledge graph of comorbidities in the elderly, which includes typical comorbidity evolution paths reported by different medical institutions; The disease interaction feature matrix of the target elderly is matched with the knowledge graph using a subgraph. When the similarity to a known high-risk comorbidity pathway is detected to exceed a threshold, an emergency consultation marker is added to the set of personalized intervention strategies.

9. The method for predicting multiple comorbidities in the elderly according to claim 8, characterized in that, The specific steps for subgraph matching of the disease interaction feature matrix of the target elderly with the knowledge graph include: Extract a set of reference nodes from the knowledge graph that match the age segments of the target elderly person; Calculate the cosine similarity between each feature dimension in the disease interaction feature matrix and the feature of the reference node; Filter the top five reference nodes by similarity to construct a local matching subgraph; The shortest pathological development path length between nodes in the local matching subgraph is analyzed and used as the criterion for judging the similarity.

10. A system for predicting multiple comorbidities in the elderly, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for predicting multiple comorbidities in the elderly as described in any one of claims 1 to 9.

Images