A management and diagnosis application system based on traditional Chinese and western medicine collaborative medical care service

Abstract

The application discloses a management and diagnosis and treatment application system based on traditional Chinese and western medicine collaborative medical and health service, and relates to the technical field of old-age health management.The system first constructs a three-dimensional digital twin correlation library including individual level, group level and risk prototype level, and each twin is calibrated in two dimensions of western medicine physiological parameters and traditional Chinese medicine syndrome element weight;then multi-modal health data of the monitored old people are collected, after cross-modal feature alignment processing of traditional Chinese and western medicine, the western medicine causal deduction link and the traditional Chinese medicine syndrome element transmission link are deduced in parallel, and the health care needs corresponding to the monitored old people and the twin correlation body are obtained;finally, the traditional Chinese and western medicine collaborative intervention scheme is matched, and the optimal scheme is selected through counterfactual simulation and pushed to the health care implementer.The system proposed in the application does not need to rely on real correlation objects for screening, can cover potential health risks, and improves the comprehensiveness and accuracy of health care demand prediction.

Description

Technical Field

[0001] This application relates to the field of elderly care and health management technology, and in particular to a management and treatment application system based on integrated traditional Chinese and Western medicine care and health services. Background Technology

[0002] As the aging population continues to deepen and the number of elderly people in my country continues to expand, the demand for elderly care services is extending from traditional disease treatment to full-cycle, comprehensive health management. Identifying the health needs of the elderly and intervening in health risks in advance is the core objective of improving the quality of elderly care services. Currently, elderly care management largely revolves around analyzing the individual health data of monitored elderly individuals. For example, Chinese patent CN119848792A introduces a correlation object analysis logic, which identifies individuals with close social connections to the monitored elderly, analyzes the secondary health risks arising from the monitored elderly's health events, and then provides collaborative health support to improve the comprehensiveness of health management.

[0003] However, the aforementioned technologies still suffer from two main problems. First, the selection of associated individuals heavily relies on a real, existing elderly population whose health status meets pre-defined criteria. However, many elderly individuals in health and wellness settings have underlying health conditions or subclinical states, resulting in a very low percentage fully meeting the health criteria and a severely insufficient screening pool. Simultaneously, a large number of relatively stable elderly individuals are not included in routine health monitoring databases, limiting the coverage of health data and leading to low matching success rates and incomplete screening. This is particularly problematic in home-based care and small-scale elderly care facilities, where there are often no matching individuals, making the association analysis mechanism difficult to implement effectively. Second, the selection criteria rely solely on the elderly's current visible health indicators, neglecting potential risk factors such as diet, lifestyle, genetic predisposition, and physical constitution. Some elderly individuals without current visible symptoms may have hidden health risks, making them highly susceptible to acute illnesses triggered by external stimuli such as emotional stress or lifestyle fluctuations. Relying solely on current health status to determine associated individuals will miss a large number of high-risk individuals, resulting in insufficient accuracy in predicting secondary health risks and hindering the proactive and targeted nature of health and wellness support. Summary of the Invention

[0004] This application provides a management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services to solve the problems in the prior art.

[0005] This application provides a management and diagnosis application system based on integrated traditional Chinese and Western medicine medical and health care services, including a digital twin construction module, a data acquisition and processing module, a health care needs projection module, and an integrated traditional Chinese and Western medicine intervention generation module; The digital twin construction module is used to build a set of digital twin associations including individual, group, and risk prototype levels. Each digital twin association in the set is labeled using a two-dimensional state structure of Western medicine psychological parameters and traditional Chinese medicine syndrome element weights. Among them, individual-level digital twin associations correspond to contacts with complete health records, group-level digital twin associations correspond to social contacts with incomplete data, and risk prototype-level digital twin associations correspond to unmonitored populations with potential health risks. The data acquisition and processing module is used to collect Western medicine modality data and traditional Chinese medicine modality data of the monitored elderly. Through the cross-modal feature alignment structure of traditional Chinese and Western medicine, the data of different modalities are mapped to a unified feature space to generate a fused feature vector. The health and wellness needs deduction module is used to perform parallel deduction based on the fused feature vector through the Western medicine causal transmission path and the traditional Chinese medicine syndrome element transmission path to obtain the first health and wellness needs information corresponding to the monitored elderly, and the second health and wellness needs information corresponding to each digital twin related entity; the deduction results of the two paths are cross-validated and the final health and wellness needs set is output. The integrated traditional Chinese and Western medicine intervention generation module is used to match integrated traditional Chinese and Western medicine intervention plans with the set of health and wellness needs. It evaluates the intervention effects of different intervention plans through counterfactual simulation of digital twins and selects the optimal plan to push to health and wellness implementers.

[0006] The management and treatment application system based on integrated traditional Chinese and Western medicine medical and health care services disclosed in this application has the following advantages: 1. Breaking through the limitations of screening real-world related objects, it enables comprehensive coverage of related risks across all dimensions. The three-tiered digital twin related object database provides complete coverage from strong to weak correlations, and from explicit health to potential risks. Individual-level digital twin related objects are adapted for accurate extrapolation in strong-correlation scenarios, while group-level digital twin related objects solve the problem of missing scattered individual data. Risk prototype-level digital twin related objects can cover potential risk populations not included in monitoring without relying on real-world individual data. It can stably conduct related risk analysis in various health and wellness scenarios, fundamentally avoiding the problems of difficult screening and incomplete matching of related objects.

[0007] 2. Integrating the TCM syndrome element system into the entire process of digital twin modeling and risk extrapolation, this approach captures overt physiological abnormalities and acute disease risks through Western medical quantitative indicators, while identifying subclinical constitutional imbalances and Qi stagnation through the evolution of TCM syndrome elements. The dual-link extrapolation mechanism cross-validates risk results, simultaneously identifying both existing disease treatment needs and preventative intervention needs. This compensates for the limitations of a single-dimensional approach in identifying potential health risks, enhancing the proactiveness and accuracy of health and wellness demand prediction.

[0008] 3. For intervention programs that combine traditional Chinese and Western medicine to address the same risk profile, the system uses counterfactual simulations of digital twins to quantitatively evaluate the intervention effects of different programs, automatically selecting the optimal synergistic program to improve the accuracy of interventions. Combined with a twin model self-calibration mechanism, the system can reverse-calibrate the twin model parameters based on real-world data, forming a continuously optimized health and wellness management system. The model's accuracy can be continuously improved with the accumulation of data during operation. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This application provides a functional module diagram of a management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] Figure 1 This application provides a functional module diagram of a management and diagnosis application system based on integrated traditional Chinese and Western medicine healthcare services. The system includes a digital twin construction module, a data acquisition and processing module, a healthcare needs projection module, and a traditional Chinese and Western medicine collaborative intervention generation module. The digital twin construction module is used to build a set of digital twin associations including individual, group, and risk prototype levels. Each digital twin association in the set is labeled using a two-dimensional state structure of Western medicine psychological parameters and traditional Chinese medicine syndrome element weights. Among them, individual-level digital twin associations correspond to contacts with complete health records, group-level digital twin associations correspond to social contacts with incomplete data, and risk prototype-level digital twin associations correspond to unmonitored populations with potential health risks. The data acquisition and processing module is used to collect Western medicine modality data and traditional Chinese medicine modality data of the monitored elderly. Through the cross-modal feature alignment structure of traditional Chinese and Western medicine, the data of different modalities are mapped to a unified feature space to generate a fused feature vector. The health and wellness needs deduction module is used to perform parallel deduction based on the fused feature vector through the Western medicine causal transmission path and the traditional Chinese medicine syndrome element transmission path to obtain the first health and wellness needs information corresponding to the monitored elderly, and the second health and wellness needs information corresponding to each digital twin related entity; the deduction results of the two paths are cross-validated and the final health and wellness needs set is output. The integrated traditional Chinese and Western medicine intervention generation module is used to match integrated traditional Chinese and Western medicine intervention plans with the set of health and wellness needs. It evaluates the intervention effects of different intervention plans through counterfactual simulation of digital twins and selects the optimal plan to push to health and wellness implementers.

[0013] For example, the individual-level digital twin associates target close contacts, such as spouses living at home or elderly people living in the same room in nursing homes, for accurate risk projection in strongly associated scenarios. The initial health status vector of each individual-level digital twin associate is composed of two dimensions: a Western medicine physiological baseline vector and a Traditional Chinese Medicine (TCM) syndrome element weight vector. The Western medicine physiological baseline vector includes seven main physiological parameters: resting heart rate, systolic blood pressure, diastolic blood pressure, fasting blood glucose, blood oxygen saturation, heart rate variability (SDNN), and body mass index. Each parameter is taken from the corresponding person's steady-state monitoring data over the past 30 days. During calibration, physical examination reports and 30 consecutive days of steady-state monitoring data from wearable devices are collected from the corresponding person. The arithmetic mean is then used as the initial value of the Western medicine physiological baseline vector after removing outliers exceeding the mean ± 3 standard deviations using the 3σ criterion. Simultaneously, the corresponding person's basic medical history and medication history are entered as auxiliary labels for parameter correction in subsequent disease risk projection stages.

[0014] The TCM syndrome element weight vector includes eight basic syndrome elements: Qi deficiency, Yang deficiency, Yin deficiency, Phlegm-dampness, Damp-heat, Blood stasis, Qi stagnation, and Special constitution. Each element has a value range of [0,1], with higher values ​​indicating a more significant bias in the corresponding syndrome element. The calibration process is based on the scores of the "Classification and Determination of TCM Constitution" standard scale, tongue diagnosis results, and pulse diagnosis results, calculated using a weighted summation syndrome element differentiation algorithm. Each syndrome element corresponds to 3 to 5 specific scale items, each item is scored from 1 to 5 points according to symptom severity. The item scores are multiplied by the contribution weight of the corresponding syndrome element, summed, and then corrected by combining the sign coefficients of tongue and pulse. Finally, the weight value of the corresponding syndrome element is obtained by max-min normalization mapping to the [0,1] interval. Taking the Qi deficiency syndrome as an example, there are four scale items: fatigue, shortness of breath and reluctance to speak, spontaneous sweating and susceptibility to colds. Each item has an equal weight. The physical signs of pale tongue with white coating and weak pulse have coefficients of 1.1 and 1.15, respectively. When matching physical signs, the corresponding coefficients are multiplied, and the weight values ​​of the Qi deficiency syndrome are obtained by normalization.

[0015] Group-level digital twins are constructed for socially connected populations with incomplete data and no comprehensive health records. They are used for batch risk extrapolation of weakly associated individuals without requiring binding to real individuals. Based on a horizontal federated learning framework, group-level digital twins are generated by aggregating anonymized health data of elderly people from multiple healthcare institutions, obtaining a group average health prototype without compromising individual privacy.

[0016] The federated averaging algorithm is used during aggregation, and the state parameter update formula for the global population prototype is:

[0017] In the formula, This is the state parameter vector of the global group prototype, that is, the initial health state vector of the group-level digital twin association; K The total number of healthcare institution nodes participating in the federal aggregation; For the first k The number of samples corresponding to the population in each health and wellness institution node; For the first k The average state parameter vector of the local population calculated from each health and wellness institution node.

[0018] The group-level digital twin associations are constructed by classifying and constructing data along three dimensions: age group, underlying disease type, and TCM constitution type. Age groups are divided into four categories: 60-64 years old, 65-74 years old, 75-84 years old, and over 85 years old. Underlying disease types are categorized into four types: no underlying disease, prehypertension, prediabetes, and stable coronary heart disease. TCM constitution types are categorized into five types: Qi deficiency, Yin deficiency, phlegm-dampness, blood stasis, and Qi stagnation. These three dimensions are combined to form 36 standard group prototypes. In practical applications, based on the age, medical history, and constitution characteristics of the social contacts of the monitored elderly, corresponding group-level digital twin associations are matched as the objects of association analysis. The state vector structure of the group-level digital twin associations is completely consistent with that of the individual-level digital twin associations, and all parameters are statistical averages of the corresponding population, representing the general health status of this population and solving the association analysis problem in scenarios where individual data is missing.

[0019] The risk prototype-level digital twin association is constructed for populations not included in health monitoring but with potential health risks. It is completely independent of real individual data and aims to cover potentially high-risk groups missed by existing screening mechanisms. The risk prototype-level digital twin association is jointly constructed based on Western medicine disease progression guidelines and traditional Chinese medicine's theory of preventive medicine, focusing on high-incidence disease types easily induced by emotional stress, corresponding to typical subclinical states and TCM imbalance syndromes. Each risk prototype corresponds to a combination of a Western medicine subclinical state and a TCM imbalance syndrome, such as "prehypertension - liver yang hyperactivity type," "stable coronary heart disease - qi deficiency and blood stasis type," and "cerebral ischemia - phlegm-dampness obstructing the orifices type," etc., constructing a total of 24 standard risk prototypes.

[0020] The state vector for each risk prototype also employs a dual-dimensional structure of Western medical psychological parameters and Traditional Chinese Medicine (TCM) syndrome element weights. The Western medical dimension sets key indicator thresholds for subclinical states based on clinical guidelines such as the *Chinese Guidelines for the Prevention and Treatment of Hypertension* and the *Chinese Guidelines for Cardiovascular Disease Rehabilitation / Secondary Prevention*. For example, prehypertension corresponds to systolic blood pressure of 120-139 mmHg and diastolic blood pressure of 80-89 mmHg. Corresponding physiological fluctuation sensitivity parameters are also set to characterize the amplitude of indicator fluctuations under emotional stress. In the TCM dimension, the weights of the main syndrome elements corresponding to the syndrome type are set to high values ​​of 0.7 to 0.8, related syndrome elements to medium values ​​of 0.3 to 0.5, and irrelevant syndrome elements to low values ​​below 0.2. Taking the liver yang hyperactivity type as an example, the weights of the corresponding qi stagnation and yin deficiency syndrome elements are set to 0.75, the blood stasis syndrome element to 0.4, and the weights of other syndrome elements to below 0.2. The parameters of the risk prototype-level digital twin correlation are all derived from publicly available clinical guidelines and TCM diagnostic standards, requiring no real individual data support, and can cover all potential populations whose diseases may be induced by emotional stress.

[0021] The data acquisition and processing module performs data acquisition through a composite acquisition port. The acquired data is divided into two categories: primary multimodal data and related scenario data of the monitored elderly. Primary multimodal data is further divided into Western medicine and Traditional Chinese Medicine (TCM) modal data. Western medicine modal data includes time-series sensor data such as heart rate, blood pressure, blood oxygen, and acceleration collected by wearable devices at a sampling frequency of 1Hz; video behavior data collected by monitoring cameras at a frame rate of 15fps; and audio data collected by audio equipment at a sampling rate of 16kHz. TCM modal data includes tongue image data collected once daily upon waking, audio data collected simultaneously with voice diagnosis, pulse data collected once daily, and data on diet, sleep patterns, and emotional state collected through daily questionnaires. Related scenario data includes video and audio recordings of health events of the monitored elderly, used to determine the severity and public nature of the events; and emotional and behavioral data and physiological sensor data of the corresponding individuals in the digital twin, used to dynamically update the real-time status of the digital twin. All data collection processes comply with privacy protection regulations. Data is collected and used only after the subjects have given their informed consent. Data transmission and storage are encrypted using the national cryptographic algorithm SM4.

[0022] For data of different modalities, corresponding feature extraction networks are used to extract high-dimensional feature vectors. For Western medicine sensor data, a one-dimensional convolutional neural network combined with a long short-term memory network is used to extract features. The input is time-series physiological data. The one-dimensional convolutional layer extracts local temporal features, and the long short-term memory network captures long-term temporal dependencies. The final output is a physiological feature vector with a dimension of 128. The network contains three one-dimensional convolutional layers with kernel sizes of 7, 5, and 3, and channel numbers of 32, 64, and 128, respectively; it also contains two long short-term memory layers with a hidden layer dimension of 64; finally, a fully connected layer outputs the feature vector, and the activation function is ReLU.

[0023] The video behavior data is processed using a 3D convolutional neural network to extract action features. The input is 16 consecutive frames of RGB images, and the output is a 256-dimensional behavior feature vector that can identify behavioral states such as falling, sitting still, walking, and emotional excitement. The backbone network adopts a C3D network structure, which contains 5 layers of 3D convolutional and pooling layers, with the top layer connected to a fully connected layer to output features.

[0024] The audio data is first processed to extract Mel-frequency cepstral coefficient features with a dimension of 40, a frame length of 25ms, and a frame shift of 10ms. Then, a Transformer encoder is used to extract contextual features, outputting an audio feature vector with a dimension of 128, which can identify features such as emotional intensity and painful groans in speech. The Transformer encoder contains two coding layers and four attention heads.

[0025] Traditional Chinese medicine (TCM) tongue features were extracted using a ResNet18 network. The input was a normalized 224×224 image of the tongue region, and the output was a 256-dimensional tongue feature vector, corresponding to TCM diagnostic features such as tongue color, tongue coating color, tongue shape, and tongue coating texture. The network was fine-tuned using a TCM tongue image dataset based on ImageNet pre-trained weights.

[0026] In the extraction of TCM acoustic features, the Mel-spectrum of the audio is first extracted, with a size of 128×128. A two-dimensional convolutional neural network is then used to extract acoustic features, outputting a 128-dimensional acoustic feature vector that corresponds to the pitch, intensity, and turbidity of the sound, among other TCM acoustic characteristics. The network consists of three two-dimensional convolutional layers with a kernel size of 3×3.

[0027] The cross-modal feature alignment structure of traditional Chinese medicine and Western medicine adopts a dual encoder-common projection architecture, which includes a Western medicine encoder, a traditional Chinese medicine encoder and a common projection layer. It maps the multimodal features of Western medicine and traditional Chinese medicine to the same common feature space, realizing the alignment and fusion of features of traditional Chinese medicine and Western medicine.

[0028] The Western medicine encoder receives three types of Western medicine features: physiological features, behavioral features, and audio features. After feature concatenation, these features are passed through two fully connected layers with hidden layer dimensions of 512 and 256, respectively, outputting a unified Western medicine feature vector with dimension 256. The Traditional Chinese Medicine (TCM) encoder receives two types of TCM features: tongue appearance features and sound diagnosis features. After feature concatenation, these features are passed through two fully connected layers with hidden layer dimensions of 384 and 256, respectively, outputting a unified TCM feature vector with dimension 256. A common projection layer projects the unified Western medicine features and the unified TCM features onto a common feature space of dimension 512, respectively, to obtain the Western medicine common features and the TCM common features. Simultaneously, a cross-modal attention mechanism is introduced to automatically learn the contribution weights of different modal features to different health states.

[0029] The formula for calculating cross-modal attention weights is:

[0030] In the formula, For the first m Attention weights for each modality feature, with the sum of all weights being 1; For the first m The association score between a modality feature and the target health status is calculated by a two-layer fully connected network. The first layer maps the single-modality feature vector into a 64-dimensional hidden vector with ReLU activation function, and the second layer outputs a scalar association score with a linear activation function. The input is the feature vector of the corresponding modality and the embedding vector of the target health status. The association strength between the modality feature and the health status is learned through the fully connected network. M The total number of modalities participating in the fusion is 5 in this embodiment, which includes 3 types of Western medicine modalities and 2 types of traditional Chinese medicine modalities; j This is a modal traversal index, used only for summation calculations.

[0031] Through the attention mechanism, the cross-modal feature alignment structure of traditional Chinese and Western medicine can automatically adjust the weights of different modalities. When identifying cardiovascular risks, the weight of electrocardiogram sensor data is automatically increased; when identifying spleen deficiency syndrome, the weight of tongue appearance features is automatically increased, thus achieving adaptive adjustment of modal weights.

[0032] The cross-modal feature alignment structure for Traditional Chinese Medicine (TCM) and Western medicine is pre-trained using a contrastive learning framework, and then its task adaptability is improved through supervised fine-tuning. The training data in the pre-training stage consists of paired Western medicine-TCM health datasets. Each sample contains Western medicine physical data and TCM diagnostic data for the same subject during the same period, and the label is the subject's health status classification.

[0033] The formula for calculating the total contrast loss is:

[0034] In the formula, Total comparative loss; For similarity loss, the cosine distance between the common features of Western medicine and the common features of traditional Chinese medicine for the same subject is calculated, with the goal of reducing the distance between similar features. The formula is as follows: ,in As a public characteristic of Western medicine, This is a public characteristic of Traditional Chinese Medicine; To achieve differential loss, the cosine distance between features of different subjects is calculated, with the goal of increasing the distance between heterogeneous features. A hard negative sample sampling strategy is used to select negative samples, with each positive sample corresponding to 3 hard negative samples. This is the balance coefficient between the two types of losses, with a value of 0.5.

[0035] Pre-training employed the Adam optimizer with an initial learning rate of 1e-4, a batch size of 32, and 50 training epochs. Training stopped when the validation set loss did not decrease for five consecutive epochs. During fine-tuning, supervised fine-tuning was performed on a labeled dataset of elderly care scenarios, with labels corresponding to the elderly care demand categories. The loss function used was cross-entropy loss. The parameters of the underlying feature extraction network were frozen during fine-tuning, and only the encoder and projection layer parameters were updated. The fine-tuning epochs consisted of 10 epochs with a learning rate of 5e-5. After fine-tuning, the model could directly output aligned fused feature vectors for subsequent risk projection.

[0036] When generating the primary health and wellness needs of the monitored elderly, the health and wellness needs deduction module first obtains single-dimensional health and wellness needs based on the multimodal fusion features of the monitored elderly through both a Western medicine classifier and a traditional Chinese medicine (TCM) syndrome differentiation tool. Then, a weighted fusion is used to obtain the primary health and wellness needs. The Western medicine classifier outputs the incidence probabilities of various diseases, covering 12 core high-incidence diseases and injury types among the elderly, such as heart attack, hypoglycemia, and fall injuries. Each element in the output vector corresponds to the incidence probability of one type of disease. The TCM syndrome differentiation tool outputs the matching degree of various syndrome types, covering 16 common TCM syndrome types among the elderly, such as Qi deficiency syndrome, blood stasis syndrome, and liver yang hyperactivity syndrome. Each element in the output vector corresponds to the matching degree of one type of syndrome.

[0037] The first health and wellness demand vector is formed by multiplying the results of Western medicine classification with their corresponding confidence weights and the results of traditional Chinese medicine diagnosis with their corresponding confidence weights, respectively. The confidence weights of various Western medicine demands are determined by the average recognition accuracy of the corresponding modality, and the confidence weights of various traditional Chinese medicine demands are determined by the average diagnosis accuracy of the corresponding modality, with values ​​ranging from [0,1]. Items in the first health and wellness demand vector with values ​​higher than a preset threshold are selected as the final first health and wellness demand information. At the same time, the severity level of each demand is assessed and divided into four levels: mild, moderate, severe, and extremely severe. The grading standard refers to the degree of injury classification in the International Statistical Classification of Diseases and Related Health Problems.

[0038] After a health event occurs in a monitored elderly person, the real-time status of the digital twin is first updated based on the event's context information, serving as the basis for subsequent inferences. The initial calculation of the health event's stimulus intensity is as follows:

[0039] In the formula, I The intensity of the stimulus to the digital twin associated with the event is denoted by , which takes the value range [0,1] and is dimensionless. s The severity coefficient for health events is 0.2 for mild, 0.5 for moderate, 0.8 for severe, and 1.0 for extremely severe. It is dimensionless. e The correlation strength coefficient is set to 1.0 for individual-level digital twin correlations, 0.6 for group-level digital twin correlations, and 0.4 for risk prototype-level digital twin correlations, and is dimensionless.

[0040] The emotional state parameters of the digital twin are updated based on stimulus intensity. In the Western medicine dimension, emotional arousal is the baseline emotional arousal plus the product of stimulus intensity and emotional sensitivity coefficient. The baseline emotional arousal ranges from [0,1], and the emotional sensitivity coefficient is determined by the baseline physiological state of the digital twin, ranging from [0.2,0.8]. The final output emotional arousal is limited to the [0,1] interval, with any value exceeding this range truncated. In the Traditional Chinese Medicine dimension, emotional disturbance is the baseline emotional stability plus the product of stimulus intensity and emotional sensitivity coefficient. The baseline emotional stability ranges from [0,1], and the emotional sensitivity coefficient is determined by the baseline syndrome state of the digital twin. Those with Qi stagnation or Qi deficiency have higher sensitivity coefficients, ranging from [0.3,0.9]. The final output emotional disturbance value is limited to the [0,1] interval, with any value exceeding this range truncated.

[0041] The Western medicine causal transmission path is based on the causal transmission logic of health events, emotional stress, physiological fluctuations, and disease risk. It deduces the Western medicine disease risk of digital twin related entities and generates a second health and wellness demand in the Western medicine dimension.

[0042] The Western medicine model employs a causal graph combined with a Transformer encoder. The causal graph defines the causal path between event stimulation, emotional arousal, physiological index fluctuations, and disease risk. The Transformer encoder extracts temporal features and nonlinear relationships. The causal graph contains four layers of nodes, from top to bottom: the event stimulation layer, the emotional arousal layer, the physiological index fluctuation layer, and the disease risk layer. The event stimulation layer includes two nodes: the severity level of the health event and the duration of event exposure. The emotional arousal layer includes two nodes: sympathetic nerve excitation and cortisol level prediction. The physiological index fluctuation layer includes four nodes: heart rate, systolic blood pressure, diastolic blood pressure, and heart rate variability. The disease risk layer includes nodes for high-incidence diseases such as acute coronary heart disease, stroke, and hypoglycemic coma. Edges between layers represent positive causal transmission, meaning that an increase in the value of an upper-layer node leads to an increase in the corresponding value of a lower-layer node. Covariant edges are set between nodes within the same layer to represent the coupling relationship between physiological indicators. The edge weights of the causal graph are initialized using clinical epidemiological data and optimized synchronously with the model during training.

[0043] The input to the Western medicine causal transmission path consists of the baseline physiological state vector, emotional arousal level, health event severity level, and medical history labels of the digital twin associated entity. The output is the incidence probability of various high-incidence diseases, i.e., the probability vector of the second health care need in Western medicine. The model contains a 3-layer Transformer encoder layer, a hidden layer dimension of 256, a multi-head attention head of 8, a feedforward network dimension of 1024, and finally outputs the incidence probability of various diseases through a fully connected layer and a Sigmoid activation function.

[0044] The training dataset for the Western medicine causal transmission pathway consists of physiological data and disease records corresponding to historical emotional stress events. Each sample includes baseline physiological data before the event, the intensity of the event stimulus, physiological changes after the event, and a disease outcome label. The data is sourced from publicly available elderly health datasets and anonymized historical data from partner elderly care institutions, with a total sample size of no less than 5000. The loss function uses a weighted binary cross-entropy loss, with the following formula:

[0045] In the formula, This represents the total cross-entropy loss. N This refers to the batch sample size. For the first n The true disease incidence label for each sample is 1 for disease incidence and 0 for no disease incidence, which is dimensionless. For the predicted first n The incidence rate of each sample, with a value range of [0,1], is dimensionless. This is the positive sample weighting coefficient, with a value of 3.0, used to address the imbalance between positive and negative samples. It is dimensionless.

[0046] Training employed the AdamW optimizer with an initial learning rate of 2e-5, a weight decay coefficient of 1e-4, a batch size of 16, and 30 training epochs. An early stopping strategy was used to prevent overfitting; training stopped if the validation set AUC did not improve for three consecutive epochs. After training, the model can predict the incidence probabilities of various diseases in the digital twin association based on the input parameters. Diseases with probabilities exceeding a preset threshold represent the second health and wellness need in the Western medicine dimension.

[0047] Based on the TCM pathogenesis logic of emotional stimulation, qi disorder, syndrome element deviation, and disease progression, the pathway of TCM syndrome element transmission and transformation is used to deduce the changes in syndrome elements and the evolution trend of syndrome types in digital twin related entities, thereby generating a second health and wellness demand in the TCM dimension.

[0048] The TCM deduction method employs a graph attention network to construct a syndrome element transmission model. Eight basic syndrome elements serve as graph nodes, and the transmission relationships of syndrome elements in TCM pathogenesis theory are used as directed edges to construct a syndrome element association graph. Specific transmission relationships include: Qi stagnation leading to blood stasis; Qi depression leading to fire; Qi deficiency leading to Yang deficiency; Yin deficiency leading to internal heat; phlegm-dampness obstructing Qi flow leading to Qi stagnation; and blood stasis aggravating Qi deficiency. Each edge corresponds to a transmission coefficient, with initial values ​​determined by the diagnostic criteria in *Internal Medicine of Traditional Chinese Medicine*, and fine-tuned based on clinical data during training.

[0049] The input to the syndrome element transfer model is the baseline syndrome element vector of the digital twin associated entity, the emotional perturbation value, and the emotional type, including fright, anger, worry, and pensiveness. The model output is the matching degree between the perturbed syndrome element vector and the corresponding syndrome type, i.e., the second health and wellness demand vector in Traditional Chinese Medicine. The model contains two graph attention layers with four attention heads and a hidden layer dimension of 64. Finally, the updated syndrome element weight vector is output through a fully connected layer, and the matching degree of each syndrome type is obtained through the syndrome type matching matrix.

[0050] The training dataset for the model consists of clinical case data on emotional-related diseases in Traditional Chinese Medicine (TCM). Each sample includes the patient's baseline syndrome elements, the type and intensity of emotional stimuli, changes in syndrome elements after onset, and the final syndrome label. The data is sourced from publicly available TCM clinical datasets and desensitized case data from collaborating TCM hospitals, with a total sample size of no less than 3000. The loss function combines mean squared error loss with classification cross-entropy loss, and the formula is as follows:

[0051] In the formula, To prove the total loss of the element transfer model; The formula for the mean squared error loss of the evidence prediction is as follows: ,in The predicted first q Item weights, For the true first q Item weights, SThe total number of evidence elements is 8 in this embodiment of the application, and the weights of all evidence elements are dimensionless values. The cross-entropy loss for certificate type classification is used to optimize the accuracy of certificate type recognition; The balance coefficient is set to 0.6, prioritizing the accuracy of the element prediction and is dimensionless.

[0052] Training employed the Adam optimizer with an initial learning rate of 1e-4, a batch size of 32, and 40 training epochs. Training ceased when the validation set accuracy failed to improve for six consecutive epochs. After training, the model could predict the direction and degree of syndrome shift in the digital twin association based on emotional stimulus parameters. Syndromes with a matching degree exceeding a preset threshold represented the second health and wellness need in the TCM dimension, corresponding to intervention needs for disease prevention.

[0053] The results of the two path deductions are cross-validated to output the final set of health and wellness needs. A many-to-many mapping relationship between Western medicine diseases and TCM syndrome types is pre-established. This mapping relationship is constructed based on the diagnostic and treatment standards of "Integrated Traditional Chinese and Western Medicine Internal Medicine." Specific correspondences include: acute coronary heart disease corresponding to Qi deficiency and blood stasis syndrome; primary hypertension corresponding to liver Yang hyperactivity syndrome; type 2 diabetes corresponding to Yin deficiency and dryness-heat syndrome; stroke corresponding to wind-phlegm-blood stasis syndrome; and chronic fatigue corresponding to spleen deficiency and dampness accumulation syndrome. Each type of Western medicine disease can correspond to 1 to 3 core TCM syndrome types, and each TCM syndrome type can also correspond to multiple types of Western medicine diseases.

[0054] If the matching degree between a certain type of Western medicine disease risk and its corresponding TCM syndrome is higher than the corresponding threshold, it is judged as a high-confidence health care need and the risk level is upgraded; if only a single path outputs risk, it is judged as a medium-confidence supplementary health care need, and the risk level remains unchanged. The final output of the second health care need includes information in three dimensions: disease or syndrome type, risk level, and confidence level, sorted from high to low risk level.

[0055] The integrated traditional Chinese and Western medicine (TCM) intervention generation module pre-builds a TCM-Western medicine intervention knowledge base. For each type of health and wellness need, it sets multiple TCM-Western medicine intervention plans, forming an intervention matrix. Each intervention plan includes two parts: a Western medicine intervention sub-plan and a TCM intervention sub-plan. The Western medicine intervention sub-plan includes monitoring measures, drug intervention, emergency treatment, and nursing points, and is divided into three levels according to intervention intensity: Level 1 corresponds to routine monitoring and basic nursing care; Level 2 corresponds to intensive monitoring and symptomatic drug intervention; and Level 3 corresponds to emergency treatment and specialized nursing care. Taking a cardiovascular risk scenario as an example, Level 1 intervention includes hourly monitoring of heart rate and blood pressure, and bed rest; Level 2 intervention includes continuous ECG monitoring, nitroglycerin on standby, and semi-recumbent rest; and Level 3 intervention includes cardiopulmonary resuscitation preparation, emergency drug infusion, and linkage with the emergency department.

[0056] The TCM intervention sub-program includes acupressure, dietary therapy, emotional regulation, moxibustion, and massage. It is also divided into three levels based on intervention intensity: Level 1 intervention provides basic emotional regulation and daily dietary recommendations; Level 2 intervention provides acupressure guidance and targeted dietary regulation; and Level 3 intervention provides professional moxibustion and massage with specific emotional intervention. Taking the liver yang hyperactivity syndrome corresponding to cardiovascular risk as an example, Level 1 intervention includes emotional calming guidance and chrysanthemum and goji berry tea; Level 2 intervention includes guidance on acupressure at Neiguan and Taichong acupoints and rose and kudzu root tea; and Level 3 intervention includes acupressure by a professional TCM practitioner, auricular acupressure, and one-on-one emotional regulation.

[0057] Each intervention plan is labeled with three attributes: intervention intensity, intervention cost, and expected effectiveness. The intervention cost is converted into a standardized score based on manpower and material consumption. The expected effectiveness is calibrated based on clinical literature data, with a value range of [0,1], and is used for subsequent plan selection. When matching plans, based on the final output of the type of health and wellness needs, risk level, and confidence level, 3 to 5 candidate collaborative plans of the corresponding level are retrieved from the knowledge base and enter the counterfactual simulation stage. This ensures plan coverage while controlling the amount of computation to meet the response requirements of real-time simulation.

[0058] Counterfactual simulation, based on a state evolution model of digital twin related entities, simulates changes in the health status of digital twin related entities under different intervention schemes, and quantitatively evaluates the intervention effects. State evolution employs differential equations, the formula of which is:

[0059] In the formula, for t A health status vector of a real-time digital twin, with each element corresponding to different physiological indicators and factor weights; This is the rate of change of the state vector over time, with the same units per time as the corresponding elements; for t The vector of intervention measures applied at any given time, with each term corresponding to the intensity of an intervention measure. When there is no intervention, the corresponding term is 0, which is dimensionless. Individualized parameters for digital twins, including physiological sensitivity, factor transfer rate, etc. f The state evolution function is composed of the trained Western medicine causal transmission path and syndrome element transmission model, which describes the change law of the state with the intervention measures, and the output dimension is consistent with the state vector.

[0060] The above equations were solved numerically using the fourth-order Runge-Kutta method, with a solution step size of 0.5 hours and a total simulation duration of 24 hours, involving 48 iterations. Each iteration calculated the rate of change of state based on the current state vector and the intervention vector, updating the state value for the next moment, ultimately outputting a complete time-series curve of state change. The specific iterative formula is as follows:

[0061] wherein, h is the solution step size, with a value of 0.5 hours; to are four intermediate slope values of the fourth-order Runge-Kutta method, with dimensions consistent with the state change rate; is the state vector at the end of the step, with dimensions consistent with the initial state vector.

[0062] The execution process of counterfactual simulation is as follows: taking the state of the digital twin associated body at the time when the health event occurs as the initial state, inputting the intervention measure vectors corresponding to different candidate intervention schemes respectively, performing iterative solution through the fourth-order Runge-Kutta method, simulating the health state change of the digital twin associated body within the next 24 hours, and calculating the risk reduction rate corresponding to each scheme. The risk reduction rate is the ratio of the final risk value after intervention to the risk value without intervention. A smaller ratio indicates a better intervention effect. The risk value is obtained by weighted calculation of the onset probability of the corresponding disease and the syndrome type matching degree, and the weight is consistent with the confidence of the health care demand.

[0063] A multi-objective optimization strategy is adopted when screening the optimal scheme to calculate the comprehensive score of each candidate scheme, and the formula is as follows:

[0064] wherein, R is the comprehensive score of the scheme. The higher the score, the better the scheme, with a value range of [0,1], which is dimensionless; E is the intervention effectiveness, that is, the risk reduction rate obtained by counterfactual simulation, with a value range of [0,1], which is dimensionless; C is the normalized intervention cost, with a value range of [0,1], which is dimensionless; is the weight coefficient of the intervention effect, with a value of 0.7, which prioritizes ensuring the intervention effect and is dimensionless.

[0065] The collaborative scheme with the highest comprehensive score is selected as the final scheme, the corresponding standardized disposal process is retrieved, and the scheme details, operation steps, precautions and expected effects are pushed together to the hand-held terminal device of the health care execution personnel, so that the execution personnel can carry out health care support on site according to the process guidance. For health care demands with a high risk level, the system synchronously pushes an early warning prompt to the health care management background, so that the management personnel can follow up the execution progress to ensure that the intervention measures are implemented on time.

[0066] In the embodiment of the present application, a twin model self-calibration module is further included. The module is a parameter iterative optimization unit supporting the system, which is used to collect real health data after the intervention is executed, compare the error between the simulated state and the real state, and reversely calibrate the individualized parameters of the digital twin associated body and the operation parameters of each deduction model, so that the system accuracy can be continuously improved with the accumulation of data during the operation process.

[0067] Following the intervention, multimodal health data from the monitored elderly and related individuals were continuously collected for 24 hours as the baseline data for calibration. From a Western medicine perspective, time-series physiological data such as heart rate, blood pressure, blood oxygen saturation, and heart rate variability were collected, with the collection frequency consistent with baseline monitoring to ensure complete alignment of data dimensions with the simulation output. From a Traditional Chinese Medicine perspective, tongue and pulse data were collected after the intervention, combined with information from daily emotional, dietary, and lifestyle questionnaires to verify the current syndrome status. All data collection adhered to privacy protection guidelines and was used solely for model parameter calibration, without disclosing individual identity information.

[0068] The data collection process synchronously records the actual implementation of the intervention plan, including the intervention start time, the actual intensity of each measure, adjustments made during implementation, and personnel operational deviations. This information is used to eliminate state errors caused by implementation deviations. If the difference between the actual implementation and the simulated preset conditions exceeds a preset range, the corresponding sample is not included in the calibration dataset to avoid invalid data interfering with the accuracy of the model parameters.

[0069] The state error vector is obtained by subtracting the actual health status vector collected within the same time window from the simulated state vector output by the digital twin. The formula is as follows:

[0070] In the formula, This is the state error vector, with the same dimension as the healthy state vector, and the dimensions of each element are consistent with the corresponding state parameters. The health status vector is the actual data collected. The health status vector obtained from the simulation of the digital twin is the same-time health status vector.

[0071] Based on the state error-based reverse calibration of the individualized parameters of the digital twin correlation, a mini-batch stochastic gradient descent method is used to update the parameters, gradually approximating the real state from the simulated state. The parameter update formula is:

[0072] In the formula, This is the individualized parameter vector before the update; This is the updated individualized parameter vector; The learning rate is 0.001 for a single calibration of an individual-level digital twin and 0.0005 for batch updates of group-level and risk prototype-level digital twins, which is dimensionless. The gradient of the state error loss with respect to the individualized parameters is used as the loss function, which is the mean squared error, i.e., the average of the squared differences between the true and simulated values ​​of each dimension of the health state vector.

[0073] Different calibration frequencies and strategies are employed for digital twins at different levels. Individual-level digital twins correspond to fixed real individuals, and a single calibration is performed after each health event intervention to gradually correct personalized parameters such as individual physiological sensitivity and evidence transfer rate, thereby improving the matching degree between the twin and the corresponding individual. Group-level and risk prototype-level digital twins are designed for general population characteristics. A batch parameter update is performed after accumulating 100 real and valid data points for the corresponding category. During the update, the basic feature framework of the prototype is preserved, and only the statistical mean and fluctuation range of the parameters are adjusted. This improves accuracy while maintaining the universality of the prototype, avoiding deviations from the general characteristics of the group due to single-sample bias.

[0074] Each inference model (i.e., the causal transmission path in Western medicine and the syndrome element transmission model in Traditional Chinese Medicine) undergoes iterative fine-tuning at a fixed cycle, set once per quarter. The fine-tuning data consists of newly added real-world event and intervention effect datasets for that quarter. The fine-tuning process employs a transfer learning strategy, retaining the low-level feature extraction network parameters obtained during the pre-training phase and only fine-tuning the parameters of the top-level fully connected and attention layers. This approach adapts the model to real-world data while avoiding catastrophic forgetting and preventing the model from losing its general feature extraction capabilities.

[0075] The validation metrics for fine-tuning fall into two categories: for the element transfer model, the AUC value of risk prediction is used as the core validation metric; for the element transfer model, the mean absolute error of element prediction is used as the core validation metric. After fine-tuning, model performance is tested on an independent validation set. When the improvement in validation metrics exceeds 2%, the online model version is updated; if the improvement does not reach the threshold, the original model version is retained, and data continues to accumulate for the next round of fine-tuning, avoiding frequent updates that could degrade system stability.

[0076] The cross-modal feature alignment structure of traditional Chinese medicine and Western medicine, the causal transmission path of Western medicine, and the syndrome element transfer model all adopt the above-mentioned fine-tuning strategy. Each model is updated independently without interference. Historical versions are retained after each model update to facilitate rapid rollback in case of anomalies, ensuring the reliability of system operation. Through the above calibration and iteration mechanism, the system can continuously optimize its accuracy during operation, adapting to the characteristics of different populations and intervention modes in various health and wellness scenarios. Long-term use will gradually improve the predictive and intervention effects.

[0077] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0078] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services, characterized in that, It includes a digital twin construction module, a data acquisition and processing module, a health and wellness needs projection module, and a traditional Chinese and Western medicine collaborative intervention generation module; The digital twin construction module is used to construct a set of digital twin associations including individual-level, group-level, and risk prototype-level digital twin associations. Each digital twin association in the set is labeled using a two-dimensional state structure of Western medicine psychological parameters and traditional Chinese medicine syndrome element weights. Among them, individual-level digital twin associations correspond to contact persons with complete health records, group-level digital twin associations correspond to social contact persons with incomplete data, and risk prototype-level digital twin associations correspond to unmonitored persons with potential health risks. The data acquisition and processing module is used to collect Western medicine modality data and traditional Chinese medicine modality data of the monitored elderly. Through the cross-modal feature alignment structure of traditional Chinese and Western medicine, the data of different modalities are mapped to a unified feature space to generate a fused feature vector. The health and wellness needs deduction module is used to perform parallel deduction based on the fused feature vector through the Western medicine causal transmission path and the traditional Chinese medicine syndrome element transmission path to obtain the first health and wellness needs information corresponding to the monitored elderly, and the second health and wellness needs information corresponding to each of the digital twin related entities; the deduction results of the two paths are cross-validated and the final health and wellness needs set is output. The integrated traditional Chinese and Western medicine intervention generation module is used to match the integrated traditional Chinese and Western medicine intervention plan corresponding to the set of health and wellness needs. The counterfactual simulation of the digital twin association is used to evaluate the intervention effect of different intervention plans, and the optimal plan is selected and pushed to the health and wellness implementers.

2. The management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services according to claim 1, characterized in that, The process by which the digital twin construction module constructs the digital twin associations at each level is as follows: The individual-level digital twin association uses the corresponding person's historical physical examination data and continuous steady-state monitoring data to calibrate the Western medicine physiological baseline, and uses the TCM constitution identification results and tongue and pulse diagnosis results to calibrate the TCM syndrome element weights, forming a two-dimensional initial state vector. The group-level digital twin association is based on a horizontal federated learning framework and uses a federated averaging algorithm to aggregate anonymized population health data from multiple elderly care institution nodes. The global group prototype's state parameter update formula is: In the formula, Let be the state parameter vector of the global group prototype. K This represents the total number of healthcare and wellness institution nodes participating in the aggregation. For the first k The number of samples from the corresponding population within each health and wellness institution node. For the first k The average state parameter vector of the local population calculated from each node of the health and wellness institution; The risk prototype-level digital twin is based on Western medicine clinical disease progression guidelines and traditional Chinese medicine's theory of disease prevention, and constructs multiple standard risk prototypes according to the combination of subclinical states and biased syndromes.

3. The management and treatment application system based on integrated traditional Chinese and Western medicine medical and health care services according to claim 1, characterized in that, The Western medicine causal transmission path of the health and wellness demand inference module adopts a structure combining a causal graph and a Transformer encoder. The causal graph has four layers of nodes from top to bottom: event stimulation layer, emotional arousal layer, physiological fluctuation layer, and disease risk layer. Positive causal transmission edges are set between layers, and covariance edges of physiological indicators are set within the same layer. The input of the Western medicine causal transmission path is the baseline physiological state, emotional arousal level, health event severity level, and medical history label of the digital twin associated entity. The output is the incidence probability of various diseases. Diseases with a probability higher than a preset threshold are the health and wellness needs in the Western medicine dimension.

4. The management and treatment application system based on integrated traditional Chinese and Western medicine medical and health care services according to claim 1, characterized in that, The TCM syndrome element transmission path of the health and wellness demand inference module is a syndrome element transmission model constructed using a graph attention network. The syndrome element transmission model uses basic syndrome elements as graph nodes and syndrome element transmission relationships in TCM pathogenesis theory as directed edges to construct a syndrome element association graph. The input of the syndrome element transmission model is the baseline syndrome element vector, emotional perturbation value, and emotional type of the digital twin association. The output is the matching degree between the perturbed syndrome element vector and the corresponding syndrome type. The syndrome type with a matching degree higher than a preset threshold is the health and wellness demand in the TCM dimension.

5. The management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services according to claim 1, characterized in that, The cross-validation process of the health and wellness demand projection module is as follows: a many-to-many mapping relationship between Western medicine diseases and TCM syndrome types is pre-established. If the matching degree between the risk of a certain type of Western medicine disease and the corresponding TCM syndrome type is higher than the corresponding threshold at the same time, it is judged as a high-confidence health and wellness demand and the risk level is upgraded. If only a single path outputs risk, it is determined to be a medium-confidence supplementary health and wellness need, and the risk level remains unchanged.

6. The management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services according to claim 1, characterized in that, The cross-modal feature alignment structure of the data acquisition and processing module adopts a dual encoder-common projection architecture, including a Western medicine encoder, a traditional Chinese medicine encoder, and a common projection layer; the cross-modal feature alignment structure of the traditional Chinese and Western medicine is pre-trained using a contrastive learning framework, and the formula for calculating the total contrastive loss is: In the formula, To compare the total losses, The similarity loss between Western medicine characteristics and traditional Chinese medicine characteristics of the same subject. For the loss of difference in characteristics between different subjects, The balancing coefficient between the two types of losses is 0.

5. After pre-training, the model is fine-tuned under supervision using a health and wellness scenario-annotated dataset, and the aligned fused feature vector is output.

7. The management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services according to claim 1, characterized in that, The counterfactual simulation of the integrated traditional Chinese and Western medicine intervention generation module uses state evolution differential equations to describe the health status changes of the digital twin associated entity. The formula for the state evolution differential equation is: In the formula, for t The health status vector of the digital twin associated at any given time. The health state vector rate of change over time for t Vector of intervention measures applied at any given time Individualized parameters for digital twin related entities, f The state evolution function is used; the fourth-order Runge-Kutta method is used to numerically solve the state evolution differential equation to simulate the time series curves of state changes under different intervention schemes.

8. The management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services according to claim 1, characterized in that, When the integrated traditional Chinese and Western medicine intervention generation module selects the optimal plan, it uses a multi-objective optimization strategy to calculate the comprehensive score of each intervention plan. The formula is as follows: In the formula, R For the comprehensive score of the intervention program, E For effective intervention, C The normalized intervention cost, The weighting coefficient for the intervention effect is set to 0.

7. The intervention plan with the highest overall score was selected as the optimal plan.

9. The management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services according to claim 1, characterized in that, It also includes a twin model self-calibration module, which is used to collect real health data after the intervention is implemented, compare the error between the simulated state and the real state, and reverse-calibrate the individualized parameters of the digital twin associated body and the operating parameters of each inference model.

10. The management and treatment application system based on integrated traditional Chinese and Western medicine medical care and rehabilitation services according to claim 9, characterized in that, The calibration process of the twin model self-calibration module is as follows: the state error vector is obtained by subtracting the real health state vector from the simulated state vector of the digital twin associated body during the same period, and the individualized parameters of the digital twin associated body are updated in reverse using the gradient descent method; wherein, the individual-level digital twin associated body performs a single calibration after each event, and the group-level and risk prototype-level digital twin associated bodies perform batch calibration after accumulating a preset amount of real data.

Images