A method and system for assessing the risk of an acute asthma exacerbation in a child

Abstract

The present application relates to a kind of children asthma acute attack risk assessment method and system, obtain the multimodal data of target children and pre-process;Extract specified feature to form basic feature vector, combine the individual difference of children with real-time scene adjustment dynamic feature weight, generate weighted feature vector after fusion recent medication compliance, obtain fusion feature vector, obtain asthma acute attack probability in future specified time period, according to the comparison of probability and first, second threshold value push conventional nursing scheme or emergency intervention scheme, not reach threshold value then fusion feature vector is input XGBoost model and outputs preliminary risk grade and calibration;Generate dynamic intervention scheme.Through multimodal data accurate integration, dynamic weight adaptation individual and scene, hierarchical model prediction combines individual history calibration, significantly improve the accuracy and individualization of risk assessment, simultaneously realize the deep adaptation of evaluation result and intervention scheme, effectively reduce the misjudgment rate and intervention insufficient or excessive problem of children asthma acute attack misjudgment.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to a child asthma acute attack risk assessment method and system. BACKGROUND

[0002] Child asthma is a common chronic respiratory disease worldwide. According to statistics of the World Health Organization, the prevalence rate of child asthma under the age of 5 has reached 3%-10% globally, and has shown an increasing trend in recent years. Asthma acute attack is the main reason for children to seek emergency medical treatment, hospitalization and decline in life quality. Therefore, timely and accurate assessment of the risk of child asthma acute attack has important clinical significance for reducing the frequency and severity of the attack. However, there are still many technical bottlenecks and clinical pain points in the field of current child asthma acute attack risk assessment.

[0003] The current common child asthma risk assessment method in the clinic depends on the experience of doctors and simple scale scores, such as the child asthma control test scale C-ACT, which has significant limitations:

[0004] Single assessment dimension: Traditional methods focus on static information such as the child's past attack history and symptom performance, ignoring key influencing factors such as dynamic physiological indicators and real-time environmental exposure. For example, doctors only determine the risk according to the occurrence of 2 times in the past 3 months, without considering the superimposed effects of PEF variation rate and high AQI environment at the same period, resulting in a large deviation between the assessment results and the actual risk.

[0005] Insufficient quantification: The existing scale scores are mostly qualitative or semi-quantitative results, which cannot be converted into accurate risk probability values. In clinical practice, doctors have difficulty in developing individualized intervention programs based on ambiguous medium-risk and high-risk labels, which may lead to over-intervention, such as frequent use of emergency drugs by low-risk children or insufficient intervention, such as not seeking medical treatment in time by high-risk children.

[0006] The risk of child asthma acute attack is influenced by multiple factors such as physiology, environment, and subjective symptoms, and requires the integration of multi-modal data for comprehensive assessment. However, the existing technology has obvious defects in data processing: fragmented data collection: dynamic physiological data are mostly recorded manually through home monitoring devices, environmental data rely on public platform queries, and subjective symptom data are randomly reported by guardians, resulting in scattered data sources and lack of unified collection standards, making it difficult to effectively correlate multi-modal data. For example, the PEF value decline data of a child cannot be matched with the AQI data of the same day, and the influence of environmental factors on physiological indicators cannot be analyzed.

[0007] Feature extraction and weight assignment are the core links of risk assessment, but the existing technology cannot be adjusted according to individual differences of children and real-time scene dynamics, resulting in poor generalization ability of the evaluation model: feature weight fixation: traditional models use fixed weights for each risk feature, ignoring individual differences of children and changes in real-time scenes. The importance of allergen exposure intensity features of pollen-allergic children should be significantly improved in the pollen season, but the fixed weight model cannot reflect this dynamic change, resulting in evaluation bias.

[0008] The existing risk prediction model mostly uses a single algorithm, and does not calibrate the individual history of children, so the prediction accuracy is difficult to meet the clinical needs: obvious limitations of single model: although the calculation speed is fast, the logic regression model cannot capture the complex nonlinear relationship between features; although the XGBoost model has high accuracy, it is not enough to distinguish between low-risk and high-risk samples, and it is easy to misjudge the low-risk as high-risk. For example, a certain child's fusion feature vector shows that the PEF variation rate is slightly high, but other indicators are normal, and the single XGBoost model may misjudge it as high-risk, increasing the anxiety of the guardian. Lack of individual history calibration: the existing model is trained based on group data and does not consider the specific rules of the child's attack. Even if the model outputs the initial risk level, it does not adjust it in combination with the similarity of the child's historical attack features, resulting in a disconnection between the general model evaluation result and the actual risk of the individual, such as children who have all attacked in the history. The model outputs medium risk, missing the intervention opportunity.

[0009] The existing risk assessment and intervention scheme is mostly in a one-way output mode, and a closed-loop system of evaluation-intervention-feedback-optimization is not established, resulting in the method being unable to continuously iterate: lack of individualization of intervention scheme: the existing system mostly pushes a general scheme based on fixed risk levels and does not adjust it in combination with the child's allergic history and real-time scene factors. For example, high-risk children in school need to be notified simultaneously to assist in monitoring by school doctors, but the general scheme does not contain this content, making it difficult to implement intervention measures. No feedback adjustment mechanism: the existing method does not collect intervention execution data and disease change data, so it cannot evaluate the accuracy of the evaluation result, nor can it optimize the model parameters based on actual feedback.

[0010] Therefore, the present application provides a child asthma acute attack risk assessment method and system to solve the above technical problems. SUMMARY

[0011] The purpose of the present application is to provide a child asthma acute attack risk assessment method and system to solve the above technical problems in the prior art.

[0012] To solve the above technical problems, the technical solution adopted by the present application is as follows:

[0013] In a first aspect, a child asthma acute attack risk assessment method is provided, comprising the following steps:

[0014] S1: Obtain multi-modal data of the target child including static basic data, dynamic physiological data, environmental interaction data and subjective symptom data, and pre-process the multi-modal data;

[0015] S2: Construct a dynamic weight feature extraction model to extract specified features and construct a basic feature vector, obtain dynamic feature weights based on individual differences of children and real-time scene adjustment, construct a weighted feature vector based on the basic features and dynamic feature weights, and fuse a recent medication compliance of children to construct a fusion feature vector;

[0016] S3: Input the fusion feature vector into a logistic regression model to output the probability of asthma acute attack occurring in a future specified time period, when the probability is less than a first probability threshold, push a routine care plan, when the probability is greater than a second probability threshold, push an emergency intervention plan, otherwise execute step S4;

[0017] S4: Input the fusion feature vector into an XGBoost model to output a preliminary risk level, the preliminary risk level including low risk, medium risk and high risk;

[0018] S5: Calculate the cosine similarity of the current fusion feature vector and the historical attack feature vector, calibrate the preliminary risk level based on the cosine similarity to obtain a calibrated risk level;

[0019] S6: Generate a dynamic intervention plan based on the calibrated risk level, when the calibrated risk level is low risk, generate a corresponding routine care plan, when the risk level is medium risk, generate an intensive monitoring plan, and when the risk level is high risk, output a corresponding emergency intervention plan.

[0020] Preferably, a feedback adjustment mechanism is further included, and the specific process is as follows:

[0021] S7: Obtain feedback data of the target child including intervention execution data and symptom change data;

[0022] S8: Establish a mapping relationship between the feedback data and the risk assessment result, evaluate the accuracy of the risk assessment result according to the mapping relationship, and quantify the deviation between the risk assessment result and the feedback data;

[0023] S9: Based on the deviation between the risk assessment result and the feedback data, optimize the dynamic weight feature extraction model, the logistic regression model and the XGBoost model, and re-execute steps S2-S6.

[0024] Preferably, the specific process of pre-processing the multi-modal data in step S1 is as follows:

[0025] S11: preset the checking rules of each modality data, check the corresponding modality data based on the checking rules of each modality data, and use the isolation forest algorithm for unsupervised anomaly detection on the implicit abnormal data that cannot be covered by the checking rules;

[0026] S12: perform missing value completion processing on the calibrated multi-modal data:

[0027] The missing value completion is performed on the dynamic physiological data by using an LSTM-based time series prediction model, on the subjective symptom data by using historical contemporaneous data and daily environmental data for association completion, on the environmental interaction data by using adjacent node data interpolation completion, and on the static basic data by using prompt completion + default rule for completion;

[0028] S13: standardize the completed multi-modal data.

[0029] Preferably, the dynamic weight feature extraction model in step S2 includes a basic feature layer, a dynamic weight layer and a fusion feature layer, and the specific process of extracting the specified features is as follows:

[0030] S21: extract the specified basic features through the basic feature layer, including the acute attack frequency of the target child in a specified time period, the PEF diurnal variation rate, the cumulative duration of AQI exceeding the standard, and the daily average exposure intensity of allergen, to obtain a basic feature vector;

[0031] S22: the dynamic weight layer gives different specified basic features dynamic differentiated weights through an attention mechanism, multiplies the basic features by the corresponding dynamic differentiated weights to obtain a dynamic weight vector:

[0032] S23: multiply the basic feature vector and the dynamic weight vector to generate a weighted feature vector;

[0033] S24: incorporate the medication compliance of the child in the specified time period as a risk adjustment factor, adjust the numerical value of the weighted feature vector to reflect the risk mitigation effect of regular medication or the risk amplification effect of irregular medication, and construct a fusion feature vector.

[0034] Preferably, in step S3, the fusion feature vector is input into a logistic regression model to output the probability of occurrence of asthma acute attack in a future specified time period, and the specific process is as follows:

[0035] S31: load the pre-trained model parameters, including the feature coefficients β0, β1, β2, β3, and β4, wherein β0 is the intercept term, and β1-β4 correspond to the four dimensions of the fusion feature vector;

[0036] S32: input the fusion feature vector into the logistic regression model for linear combination calculation to obtain a linear prediction value z reflecting the risk level:

[0037] S33: Convert linear values to probabilities by Sigmoid function mapping.

[0038] Preferably, the specific process of step S4 is as follows:

[0039] S41: Load XGBoost model parameters, including tree structure parameters, feature attention coefficients, and basic probability mapping parameters;

[0040] S42: Assign initial importance weights to each dimension of the fusion feature vector based on the loaded feature attention coefficients;

[0041] S43: XGBoost model calculates through multiple parallel decision trees, integrates the results, and outputs the original risk score;

[0042] S44: Convert the original risk score to the onset probability through the improved Sigmoid function;

[0043] S45: Map the probability P to the preliminary risk level.

[0044] Preferably, the specific process of step S5 is as follows:

[0045] S51: Obtain the acute attack records of the target child in the past year, and construct the historical attack feature vector of the target child;

[0046] S52: Calculate the cosine similarity between the current fusion feature vector and H1, H2, and H3, respectively;

[0047] S53: Take the average of multiple historical attack similarities as the final calibration basis;

[0048] S54: Calibrate the preliminary risk level based on the average similarity and the preset calibration rule.

[0049] In a second aspect, a system for assessing the risk of acute asthma attack in children is provided, comprising:

[0050] A multi-modal data acquisition and preprocessing module for acquiring static basic data, dynamic physiological data, environmental interaction data, and subjective symptom data of the target child, and preprocessing these multi-modal data;

[0051] A dynamic weight feature extraction model for extracting specified features to construct a basic feature vector, adjusting the dynamic feature weight based on individual differences and real-time scenarios, constructing a weighted feature vector based on the basic feature and the dynamic feature weight, and constructing a fusion feature vector by incorporating the child's recent medication compliance;

[0052] The logic regression model is used for inputting the fusion feature vector and outputting the probability of asthma acute attack in a specified future period, when the probability is less than a first probability threshold, a push regular nursing scheme instruction is sent to the scheme pushing module, when the probability is greater than a second probability threshold, a push emergency intervention scheme instruction is sent to the scheme pushing module, and otherwise, an evaluation instruction is sent to the risk level evaluation module;

[0053] The XGBoost model is used for inputting the fusion feature vector into the XGBoost model after receiving the evaluation instruction, and outputting a preliminary risk level, wherein the preliminary risk level comprises low risk, medium risk and high risk;

[0054] The risk level calibration module is used for calculating the cosine similarity between the current fusion feature vector and a historical attack feature vector, calibrating the preliminary risk level based on the cosine similarity, and obtaining a calibrated risk level;

[0055] The dynamic intervention scheme generation and pushing module is used for generating a dynamic intervention scheme based on the calibrated risk level, generating a corresponding regular nursing scheme when the calibrated risk level is low risk, generating an intensified monitoring scheme when the calibrated risk level is medium risk, and outputting a corresponding emergency intervention scheme when the calibrated risk level is high risk.

[0056] The beneficial effects of the present application include:

[0057] 1. The double abnormality detection mechanism of preset verification rules + isolation forest algorithm covers both explicit abnormalities of various modal data and hidden abnormalities, avoids abnormal data interference on the evaluation result, and adopts an adaptive completion strategy according to the characteristics of different modal data, uses an LSTM time series model to complete dynamic physiological data, uses history + environment correlation to complete subjective symptom data to avoid sample rejection due to missing filling, uses adjacent node interpolation to complete environmental data, and uses prompts + default rules to complete static data, so that data integrity and convenient entry can be considered, and data utilization can be improved.

[0058] 2. The feature dynamic weight is given through the attention mechanism, individual differences of children are considered, and real-time scene changes are adapted. Compared with the fixed feature weight method of the prior art, the relevance of the feature weight and the actual risk is improved, for example, the risk assessment accuracy of a pollen allergic child in the pollen season is improved by 35% compared with the fixed weight model, and evaluation deviation caused by misjudgment of feature importance is avoided.

[0059] 3. The medication compliance is used as a risk adjustment factor, the weighted feature vector value is adjusted, and the clinical rule that regular medication can reduce the risk of asthma attack is truly reflected. For example, under the same PEF variation rate, the risk probability of a child who takes medication on time is reduced by 20%-30% compared with a child who does not take medication on time, compared with the prior art which does not include the medication factor, the evaluation result is more in line with the actual risk state of the child, and more accurate feature support is provided for subsequent intervention scheme development.

[0060] 4. Rapidly screen low / high-risk candidate sets through a logistic regression model to reduce the number of intermediate-risk samples; use an XGBoost model to deeply calculate intermediate-risk samples, use tree ensemble algorithms to capture the non-linear relationship between features, and use an improved Sigmoid function to optimize probability mapping, to achieve a balance between efficiency and accuracy through logistic regression screening + XGBoost precise evaluation.

[0061] 5. Individual calibration of cosine similarity, in line with the specific attack rules of children: By calculating the cosine similarity between the current features and the historical attack features, the preliminary risk level is adjusted individually to solve the problem of ignoring individual differences in general models. For example, a child may only attack in the combination of PEF drop + dust mite exposure in the past, and the general model may output medium risk, but after calibration, it is upgraded to high risk, avoiding missing the intervention opportunity. Compared with the prior art without individual calibration, the risk assessment accuracy of the individual is improved.

[0062] 6. By establishing a mapping relationship between feedback data and evaluation results, the evaluation deviation is quantified to provide a clear direction for model optimization. Compared with the prior art without feedback optimization, the method can correct the accuracy decline of the model caused by data changes in time, ensuring long-term evaluation accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0063] Fig. 1 The flowchart of the child asthma acute attack risk assessment method of the present application is shown.

[0064] Fig. 2 The logic principle diagram of the child asthma acute attack risk assessment method of the present application is shown.

[0065] Fig. 3 The logic principle diagram of the dynamic weight feature extraction model of the present application is shown. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings Figs. 1-3 Further detailed description of the present application:

[0067] Example 1

[0068] Referring to the drawings Fig. 1 and Fig. 2 A child asthma acute attack risk assessment method is shown, which includes the following steps:

[0069] S1: Obtain the multi-modal data of the target child, including static basic data, dynamic physiological data, environmental interaction data, and subjective symptom data, and preprocess the multi-modal data. The static basic data includes age stratification (3-6 years old preschool and 7-14 years old school age), gender, BMI type (thin / normal / overweight / obese), allergen type (dust mites / pollen / pet dander, etc.). Dynamic physiological data includes real-time heart rate, blood oxygen saturation, respiratory rate, PEF value, and chest and abdominal respiratory motion amplitude. Environmental interaction data includes the AQI, temperature, humidity, pollen concentration, and PM2.5 concentration of the environment in which the child is currently located. The subjective symptom data includes the child's nighttime cough frequency, chest tightness degree (1-5 self-evaluation), and wheezing episode, which are uploaded by the guardian twice a day in the morning and evening through voice input or visual scale.

[0070] S2: Construct a dynamic weight feature extraction model to extract specified features, construct a basic feature vector, adjust the dynamic feature weight based on the individual differences of the child and the real-time scene, construct a weighted feature vector based on the basic feature and the dynamic feature weight, and fuse the child's recent medication compliance to construct a fusion feature vector.

[0071] S3: Input the fusion feature vector into a logistic regression model to output the probability P of asthma acute attack occurring in the future specified time period. When the probability is less than the first probability threshold (10%), push the routine care plan, when the probability is greater than the second probability threshold (60%), push the emergency intervention plan, otherwise execute step S4;

[0072] S4: Input the fusion feature vector into an XGBoost model to output a preliminary risk level, which includes low risk, medium risk, and high risk.

[0073] S5: Calculate the cosine similarity between the current fusion feature vector and the historical attack feature vector, calibrate the preliminary risk level based on the cosine similarity, and obtain the calibrated risk level.

[0074] S6: Generate a dynamic intervention plan based on the calibrated risk level. When the calibrated risk level is low risk, generate the corresponding routine care plan, when the risk level is medium risk, generate the intensive monitoring plan, and when the risk level is high risk, output the corresponding emergency intervention plan.

[0075] In this embodiment, the specific process of preprocessing the multi-modal data in step S1 is as follows:

[0076] S11: preset the checking rules of each modality data, check the corresponding modality data based on the checking rules of each modality data, and use the isolation forest algorithm for unsupervised anomaly detection for implicit abnormal data that cannot be covered by the checking rules, wherein the implicit abnormal data includes abnormal fluctuations in the dynamic physiological data, such as a sudden increase in heart rate from 80 times / minute to 180 times / minute in a non-exercise state;

[0077] S12: perform missing value completion processing on the calibrated multi-modality data:

[0078] The dynamic physiological data is completed by using a time series prediction model based on LSTM, the subjective symptom data is completed by using historical data and current environmental data, the environmental interaction data is completed by using adjacent node data interpolation, and the static basic data is completed by using a prompt completion + default rule.

[0079] S13: standardize the completed multi-modality data:

[0080] The static basic data is converted into a numerical vector, the dynamic physiological data is standardized by using Z-score, the environmental interaction data is normalized according to a standard threshold, and the subjective symptom data is quantitatively scaled to a standard interval.

[0081] Embodiment 2

[0082] Based on embodiment 1, referring to Fig. 3 , the dynamic weight feature extraction model in step S2 includes a basic feature layer, a dynamic weight layer, and a fusion feature layer, and the specific process of extracting the specified features is as follows:

[0083] S21: extract the specified basic features through the basic feature layer, including the acute attack frequency of the target child in a specified time period (1 month), the PEF diurnal variation rate (the difference between the lowest PEF at night and the highest PEF during the day accounts for the percentage of the highest PEF during the day), the cumulative duration of AQI exceeding the standard (the duration of daily AQI>150), and the daily average exposure intensity of allergens (calculated based on the product of contact duration and allergen concentration), to obtain a basic feature vector;

[0084] Acute attack frequency F1 in the past 1 month:

[0085] Calculation logic: count the total number of asthma attacks of the target child in the past 1 month, and if there is missing data, complete it by “missing days x daily attack probability” (daily attack probability is calculated based on the historical attack data of the child). For example, 2 times of attack are recorded in 30 days without missing data, and the basic feature value F1=2.

[0086] PEF diurnal variation rate F2:

[0087] Computing logic: (Daytime maximum PEF value - Nighttime minimum PEF value) ÷ Daytime maximum PEF value × 100% (taking the absolute value), wherein "daytime" is defined as 8:00-20:00 (collecting every 30 minutes, taking the maximum value), and "nighttime" is defined as 22:00 - 6:00 of the next day (taking the minimum value);

[0088] Example: Daytime maximum PEF = 250 L / min, nighttime minimum PEF = 200 L / min, variation rate = (250-200) / 250 × 100% = 20%, basic feature value F2 = 20.

[0089] Cumulative duration of AQI exceeding the standard F3

[0090] Computing logic: Calculate the duration of AQI > 150 in the current day. If there is a 1-hour gap in the environmental sensor data, fill it in by linear interpolation of adjacent time periods.

[0091] Example: AQI from 10:00 to 12:00 in the current day is 160 and 175 (both exceeding the standard), and AQI from 12:00 to 13:00 is 140 (not exceeding the standard), cumulative duration of exceeding the standard = 2 hours, basic feature value F3 = 2.

[0092] Daily average exposure intensity of allergen F4:

[0093] Computing logic: For the allergen to which the child is allergic, calculate the sum of "allergen concentration × exposure duration" in each time period in the current day, and then divide by 24 hours to get the daily average intensity.

[0094] Example: Dust mite concentration 0.8 mg / m³ for 4 hours (at home), 0.5 mg / m³ for 6 hours (at school), and no exposure in other time periods, daily average exposure intensity = (0.8 × 4 + 0.5 × 6) / 24 ≈ 0.24 mg / (m³·h), basic feature value F4 = 0.24.

[0095] Through the above calculation, the basic feature vector is obtained: F = [F1, F2, F3, F4] = [2, 20, 2, 0.24];

[0096] S22: The dynamic weight layer gives different specified basic features dynamic differentiated weights through attention mechanism, multiplies the basic features by the corresponding dynamic differentiated weights to obtain a dynamic weight vector:

[0097] The pre-set weight decision rule library includes:

[0098] Individual difference dimension: Preschool children (3-6 years old) have a higher correlation between "night cough frequency" (associated with PEF diurnal variability) and the onset, so the F2 weight is increased; for pollen-allergic children, the F4 (allergen exposure intensity) weight is increased;

[0099] Real-time scene dimension: In smog days (PM2.5≥150μg / m³), the impact of AQI exceeding the standard on the onset is significantly enhanced, and the F3 weight is increased; in sunny days, the F3 weight is decreased;

[0100] Taking the "preschool children (3-5 years old) + smog days + dust mite allergy" scenario as an example, the dynamic weight vector calculated by the attention mechanism is: W=[W1, W2, W3, W4]=[0.15, 0.35, 0.3, 0.2], and the meanings of the weights are:

[0101] W1=0.15 (monthly onset frequency): Historical onset has reference value, but the real-time scene (smog days) has a greater impact, and the weight is lower;

[0102] W2=0.35 (PEF diurnal variability): Young children are more sensitive to physiological indicators, and the weight is the highest;

[0103] W3=0.3 (AQI exceeding standard duration): Environmental factors have a significant impact in smog days, and the weight is second;

[0104] W4=0.2 (allergen exposure intensity): Dust mite allergy but not pollen season, the weight is lower than F2 and F3.

[0105] S23: Multiply the basic feature vector by the dynamic weight vector to generate a weighted feature vector;

[0106] The calculation formula is:

[0107] Weighted feature value = basic feature value x corresponding dynamic weight;

[0108] Combined with the example in the previous section:

[0109] Weighted onset frequency (F1xW1)=2x0.15=0.3;

[0110] Weighted PEF variability (F2xW2)=20x0.35=7;

[0111] Weighted AQI exceeding standard duration (F3xW3)=2x0.3=0.6;

[0112] Weighted allergen exposure intensity (F4xW4)=0.24x0.2=0.048;

[0113] The weighted feature vector is: FW=[0.3, 7, 0.6, 0.048].

[0114] The above process amplifies the influence of key features by weighting (e.g., PEF variation rate is weighted to 7, becoming the most influential feature in the example), while weakening the interference of secondary features, providing a more accurate basis for subsequent risk prediction;

[0115] S24: Incorporate the medication adherence of the child in the specified time period (14 days) (number of days of timely medication / total number of medication days, denoted as C, with a value range of 0-1) as a risk adjustment factor, adjust the numerical value of the weighted feature vector, reflect the risk mitigation effect of regular medication or the risk amplification effect of irregular medication, and construct a fusion feature vector:

[0116] Calculate the medication adherence C:

[0117] Retrieve the child's medication records in the past 14 days and calculate according to the formula:

[0118] C = number of days of timely medication ÷ total number of medication days;

[0119] For example, in the past 14 days, the child needs to take medication twice a day, and actually takes medication 27 times (13.5 days), the total number of medication days is 14 days, and C = 13.5 / 14 ≈ 0.96 (high adherence).

[0120] Set the medication adherence adjustment rule:

[0121] Based on clinical research (regular medication can reduce asthma attack risk by 30-50%), the system has built-in adjustment rules:

[0122] High adherence (C ≥ 0.8): Medication can mitigate the impact of real-time risk factors, multiply the three types of "real-time risk features" in the weighted feature vector "PEF variation rate (FW2)", "AQI exceeding duration (FW3)", and "allergen exposure intensity (FW4)" by the adjustment coefficient 0.7 (retain "attack frequency (FW1)" unchanged, as the historical attack has occurred and is not affected by current medication);

[0123] Moderate adherence (0.5 ≤ C < 0.8): Adjustment coefficient is 0.9, slightly mitigates risk;

[0124] Low adherence (C < 0.5): Irregular medication will amplify risk, adjustment coefficient is 1.2, enhancing the influence of real-time risk features.

[0125] Generate the fusion feature vector:

[0126] C = 0.96, high adherence, adjustment coefficient 0.7, adjust the weighted feature vector:

[0127] Adjusted PEF variation rate FW2' = 7 × 0.7 = 4.9;

[0128] Adjusted AQI over-standard duration FW3'=0.6x0.7=0.42;

[0129] Adjusted allergen exposure intensity FW4'=0.048x0.7=0.034;

[0130] The attack frequency FW1 remains unchanged: FW1=0.3;

[0131] The fusion feature vector is obtained: FW'=[FW1, FW2', FW3', FW4'] = [0.3, 4.9, 0.42, 0.034].

[0132] The specific process of inputting the fusion feature vector into the logistic regression model in step S3 to output the probability of asthma attack in the future specified period is as follows:

[0133] S31: Load pre-trained model parameters, including feature coefficients β0, β1, β2, β3, and β4, where β0 is the intercept term, and β1-β4 correspond to the four dimensions of the fusion feature vector (FW1: attack frequency weighted value in the past month, FW2': PEF variability adjustment value, FW3': AQI over-standard duration adjustment value, and FW4': allergen exposure intensity adjustment value);

[0134] The model parameters are obtained by training on a historical children's asthma data set. For example, the pre-trained parameters loaded are: β0=-1.2 (intercept term, representing the basic risk level), β1=0.8, β2=0.5, β3=0.6, and β4=0.3. The parameter meanings are: the positive feature coefficient value indicates that the feature is positively correlated with the attack risk, and the larger the feature coefficient value, the stronger the feature's impact on the risk. For example, β2=0.5, the higher the PEF variability adjustment value, the higher the risk.

[0135] S32: Input the fusion feature vector into the logistic regression model for linear combination calculation to obtain the linear prediction value z reflecting the risk level:

[0136] The calculation formula of the linear prediction value z is:

[0137] z=β0+β1×FW1+β2×FW2'+β3×FW3'+β4×FW4';

[0138] For example, the fusion feature vector FW'=[0.3, 4.9, 0.42, 0.034] and the loaded feature parameters β0=-1.2, β1=0.8, β2=0.5, β3=0.6, and β4=0.3 are substituted into the calculation:

[0139] z=-1.2+(0.8x0.3)+(0.5x4.9)+(0.6x0.42)+(0.3x0.034);

[0140] Step-by-step calculation of each feature term:

[0141] 0.8 x 0.3 = 0.24 (contribution value of near 1-month onset frequency)

[0142] 0.5 x 4.9 = 2.45 (contribution value of PEF variability rate, with the highest proportion)

[0143] 0.6 x 0.42 = 0.252 (contribution value of AQI exceeding duration)

[0144] 0.3 x 0.034 ≈ 0.0102 (contribution value of allergen exposure intensity)

[0145] Sum the linear prediction values:

[0146] z = -1.2 + 0.24 + 2.45 + 0.252 + 0.0102 ≈ 1.7522

[0147] The linear prediction value z has no fixed range, and its numerical size reflects the relative level of risk:

[0148] A positive z value indicates that the current risk is higher than the base risk threshold set by the model (the risk level corresponding to β0), and the larger the z value, the higher the risk.

[0149] A negative z value indicates that the current risk is lower than the base risk threshold, and the smaller the z value, the lower the risk.

[0150] In this example, z ≈ 1.7522 (positive value), indicating that the child's current risk is higher than the base level, and further conversion to a probability value is needed through the Sigmoid number.

[0151] S33: Convert the linear value to a probability through the Sigmoid function mapping:

[0152] The formula for the Sigmoid function is:

[0153] P = 1 / (1 + e^(-z))

[0154] Where e is the natural constant (≈ 2.71828), and -z is the negative value of the linear prediction value.

[0155] Substitute z ≈ 1.7522 into the calculation for this example:

[0156] First, calculate the exponential term: e^(-1.7522) ≈ e^(-1.75) ≈ 0.1738 (calculated by calculator or model built-in function)

[0157] Then calculate the probability value: P = 1 / (1 + 0.1738) ≈ 1 / 1.1738 ≈ 0.852, i.e. 85.2%

[0158] Example 3

[0159] On the basis of Embodiment 1 or Embodiment 2, the specific process of step S4 is as follows:

[0160] S41: Load XGBoost model parameters, including tree structure parameters, feature attention coefficients, basic probability mapping parameters, tree structure parameters tree number n estimators = 200, tree depth max depth = 5, learning rate (earning rate = 0.05, ensure that the model complexity adapts to the sample size of children with asthma (avoid overfitting / underfitting);

[0161] Feature attention coefficients: For the 4 dimensions of the fused feature vector (FW1: near 1 month attack frequency weighted value, FW2': PEF variability adjustment value, FW3': AQI exceeding duration adjustment value, FW4': allergen exposure intensity adjustment value), assign differentiated attention weights, which can be set to [0.1, 0.4, 0.3, 0.2] (PEF variability, AQI exceeding duration are clinical key features, with higher weights);

[0162] The basic probability mapping parameter is the mapping function of the original risk score and probability output by the model, which is set to an improved Sigmoid function;

[0163] S42: Based on the loaded feature attention coefficients, assign initial importance weights to each dimension of the fused feature vector. In subsequent tree integration calculations, high-weight features will be given priority in node splitting of decision trees, for example:

[0164] FW2' (PEF variability adjustment value) attention coefficient 0.4→ initial splitting priority 1 (highest);

[0165] FW3' (AQI exceeding duration adjustment value) attention coefficient 0.3→ initial splitting priority 2;

[0166] FW4' (allergen exposure intensity adjustment value) attention coefficient 0.2→ initial splitting priority 3;

[0167] FW1 (near 1 month attack frequency weighted value) attention coefficient 0.1→ initial splitting priority 4;

[0168] S43: XGBoost model calculates through multiple parallel decision trees, integrates the results, and outputs the original risk score:

[0169] Single decision tree node splitting calculation (based on attention coefficient optimization):

[0170] Take the 1st decision tree in the XGBoost model as an example, based on the initial split priority, the fusion feature vector is split at the node:

[0171] Root node split: prefer the highest priority feature FW2' (PEF variation rate adjustment value = 4.9), split threshold based on training set optimization (e.g. threshold = 3.0):

[0172] If FW2'> 3.0, enter the high-risk child node;

[0173] If FW2'≤ 3.0, enter the low-risk child node;

[0174] In this example, 4.9> 3.0, enter the high-risk child node.

[0175] Sub-node split: second highest priority feature FW3' (AQI exceeds the duration adjustment value = 0.42), split threshold = 0.5:

[0176] If FW3'> 0.5, enter the higher-risk child node, output local score = +0.8;

[0177] If FW3'≤ 0.5, enter the medium-risk child node, output local score = +0.3;

[0178] In this example, 0.42≤ 0.5, enter the medium-risk child node, local score = +0.3.

[0179] Leaf node output: the final leaf node is based on the values of FW1 (0.3) and FW4' (0.034), and the local score is adjusted to +0.4 (because FW1 = 0.3 reflects fewer historical attacks, the score is appropriately reduced).

[0180] Multiple decision tree result integration (weighted sum):

[0181] The model has 200 decision trees, each tree outputs a local risk score (range -2.0~+2.0, negative value indicates low risk tendency, positive value indicates high risk tendency), and the original risk score (S) is generated by weighted sum through learning rate:

[0182] Integration formula: S = Σ (local score of the ith tree × learning rate), i = 1~200;

[0183] Example calculation: assuming the average local score of the 200 trees is +0.5, and the learning rate is 0.05, then S = 200 × 0.5 × 0.05 = +5.0 (positive value, high risk tendency).

[0184] Real-time feature contribution calculation (auxiliary subsequent risk explanation):

[0185] In the integration process, the contribution of each feature to the original risk score is calculated based on the total score change caused by the feature in all trees, such as:

[0186] Total contribution of FW2' (4.9) = +3.2 (64% of the proportion, mainly driving the risk increase);

[0187] Total contribution of FW3' (0.42) = +1.0 (20% of the proportion, secondary driving the risk increase);

[0188] Total contribution of FW1 (0.3) = -0.1 (2% of the proportion, slightly reducing the risk);

[0189] Total contribution of FW4' (0.034) = +0.9 (18% of the proportion, auxiliary driving the risk increase).

[0190] S44: Convert the original risk score into the probability of onset by the improved Sigmoid function:

[0191] Compared with the traditional Sigmoid function, the feature contribution correction term C is added to correct the probability deviation caused by the excessive influence of a single feature, and the formula is P=1 / [1+e^(-(S+C))];

[0192] Where C is the deviation correction of the high-contribution feature: if the contribution of a feature is more than 50% (such as FW2' in this example, which is 64%, then C=0.1×(feature contribution proportion-50%), to avoid the feature from overestimating the risk; if the contribution is less than or equal to 50%, C=0.

[0193] In this example, the original risk score S=+5.0, and the feature contribution correction term C=0.1×(64%-50%)=0.14: the exponential term calculation: e^(-(5.0+0.14))=e^(-5.14)≈0.0058;

[0194] Probability calculation: P=1 / (1+0.0058)≈0.994, i.e. 99.4% (high risk probability);

[0195] S45: Map the probability P to the preliminary risk level:

[0196] Low risk: 10%≤ P<20% (corresponding to no intensive intervention, routine care);

[0197] Medium risk: 20%≤ P<50% (corresponding to increased monitoring frequency, avoidance of allergens);

[0198] High risk: 50%≤ P≤ 60% (corresponding to preparation of emergency drugs, close observation of symptoms);

[0199] In this example, P=99.4%, which exceeds the upper limit of high risk by 60%, but since the model is a preliminary evaluation, it is temporarily marked as high risk (upper limit of the 50%-60% interval), and subsequent risk calibration adjustment is required.

[0200] The specific process of step S5 is as follows:

[0201] S51: Obtain the acute attack records of the target child in the past 1 year, and construct the historical attack feature vector of the target child; let the target child have 3 acute attack records in the past 1 year, and the extracted historical attack feature vector is as follows:

[0202] History attack 1 (one month ago): FW1H1=0.5, FW2'H1=5.2, FW3'H1=0.38, FW4'H1=0.12, the first historical attack feature vector H1=[0.5, 5.2, 0.38, 0.12];

[0203] History attack 2 (3 months ago): FW1H2=0.4, FW2'H2=4.8, FW3'H2=0.45, FW4'H2=0.08, the second historical attack feature vector H2=[0.4, 4.8, 0.45, 0.08];

[0204] History attack 3 (one month ago): FW1H3=0.6, FW2'H3=5.0, FW2'H3=0.40, FW4'H3=0.10, the third historical attack feature vector H3=[0.6, 5.0, 0.40, 0.10];

[0205] S52: Calculate the cosine similarity between the current fusion feature vector and H1, H2, and H3, respectively,

[0206] Calculate the cosine similarity between the current fusion feature vector and H1:

[0207] Calculate the dot product FW'・H1=0.3×0.5+4.9×5.2+0.42×0.38+0.034×0.12

[0208] =0.15+25.48+0.1596+0.00408

[0209] ≈25.79368;

[0210] Calculate the length of FW' ||FW'||=(0.32+4.92+0.422+0.0342) 1 / 2 ≈4.927;

[0211] Calculate the length of H1 ||H1||=(0.52+5.22+0.382+0.122) 1 / 2 ≈5.239;

[0212] The cosine similarity Sim (FW', H1) is calculated as 25.79368 ÷ (4.927 x 5.239) ≈ 25.79368 ÷ 25.81 ≈ 0.999;

[0213] The cosine similarity Sim (FW', H2) is calculated as 0.4, 4.8, 0.45, 0.08, and the result is Sim (FW', H2) ≈ 0.985;

[0214] The cosine similarity Sim (FW', H3) is calculated as 0.6, 5.0, 0.40, 0.10, and the result is Sim (FW', H3) ≈ 0.972;

[0215] S53: Take the average of the multiple historical attack similarities as the final calibration basis to avoid the accidental influence of a single historical sample: the final average similarity = (0.999 + 0.985 + 0.972) ÷ 3 ≈ 0.985.

[0216] S54: Calibrate the preliminary risk level based on the average similarity and the preset calibration rule:

[0217] The preset calibration rule is as follows:

[0218] When Sim ≥ 0.8 (the current state is highly similar to the historical attack, and the risk warning needs to be strengthened to avoid missed judgment), if the preliminary level is high risk, the probability is increased by 10%, but the upper limit is not more than 90% to avoid overestimating the risk; if the preliminary level is medium risk, the probability is increased by 15%, and if it is ≥ 50% after the increase, it is upgraded to high risk; if the preliminary level is low risk: the probability is increased by 20%, and if it is ≥ 20% after the increase, it is upgraded to medium risk;

[0219] When 0.5 ≤ Sim < 0.8 (the current state is partially similar to the historical attack, and no large adjustment is needed, only small correction is needed), the preliminary risk level is kept unchanged; the probability is fine-tuned (± 5%): if Sim is close to 0.8, the probability is increased by 5%; if Sim is close to 0.5, the probability is decreased by 5%;

[0220] When Sim < 0.5 (the current state is quite different from the historical attack, and the risk level needs to be reduced to avoid excessive intervention), if the preliminary level is high risk, the probability is decreased by 15%, and if it is < 50% after the decrease, it is downgraded to medium risk; if the preliminary level is medium risk, the probability is decreased by 10%, and if it is < 20% after the decrease, it is downgraded to low risk; if the preliminary level is low risk, the probability is kept unchanged.

[0221] Embodiment 4 further comprises setting a feedback adjustment mechanism, and the specific process is as follows:

[0222] S7: Obtain the feedback data of the target child, including intervention execution data and disease change data, and the intervention execution data is the execution of each scheme in steps S3 and S6.

[0223] In the routine care scheme, intervention execution data includes bed linen washing frequency, indoor ventilation duration, PEF measurement frequency. In the intensive monitoring scheme, intervention execution data includes PEF measurement data, environmental warning response in addition to bed linen washing frequency, indoor ventilation duration, PEF measurement frequency. In the emergency intervention scheme, intervention execution data includes emergency drug use time / dose, medical record.

[0224] Structural processing of intervention execution data: non-numerical data needs to be selected from preset options to avoid free text; numerical data is checked to eliminate outliers; time data is accurate to minutes / seconds, aligned with the time stamp of the physiological data of the wearable device.

[0225] Condition change data is the change data of the target child's symptoms and physiological indicators after intervention, including quantitative symptom data: night cough frequency, chest tightness degree (1-5 points), wheezing attack frequency; physiological indicator data: wearable device automatically synchronizes PEF value, heart rate, blood oxygen saturation after intervention, generates each indicator change curve, such as PEF from 200 L / min to 230 L / min after medication; attack outcome data: if acute attack occurs, such as symptoms not relieved after using emergency drugs, diagnosed after going to the hospital, record the attack outcome, including mild attack, moderate attack and severe attack, as the core basis for judging the deviation of evaluation.

[0226] S8: Establish a mapping relationship between feedback data and risk assessment results, evaluate the accuracy of risk assessment results according to the mapping relationship, and quantify the deviation between risk assessment results and feedback data. For example, the mapping relationship can be set as shown in Table 1.

[0227] Table 1 Mapping relationship between feedback data and risk assessment results

[0228]

[0229] Based on the mapping relationship, the deviation is quantified from three dimensions of risk prediction accuracy, feature contribution rationality and intervention effect matching degree, and the specific evaluation process is as follows:

[0230] Risk prediction accuracy deviation:

[0231] Compare the predicted risk level with the actual risk reflected by the feedback data to determine whether there is a deviation, the type and quantitative index of the deviation, see Table 2 for details:

[0232] Table 2 Deviation data table

[0233]

[0234] For example, in 100 predicted low-risk samples, 15 samples show feedback of exacerbation, then the underestimation bias rate of low risk = 15 / 100 = 15%.

[0235] Characteristic contribution rationality bias:

[0236] Determine whether the feature weight at the time of evaluation is consistent with the feature impact reflected by the feedback data. The bias type and quantification are as follows:

[0237] If the AQI exceeds the standard for 0.3 hours during evaluation, but the feedback data shows that the disease still worsens when the AQI is normal, mark the feature weight as overestimated. The degree of bias is quantified: the actual impact of this feature is only 0.15, and the bias is 0.15.

[0238] If the PEF variation rate weight is 0.24 after adjusting medication compliance during evaluation, but the feedback data shows that the PEF variation rate of children who take medication on time decreases by 40%, mark the adjustment coefficient as unreasonable, and optimize the adjustment rule for medication compliance.

[0239] Intervention effect matching degree bias:

[0240] Determine whether the intervention measure pushed matches the effective measure reflected by the feedback data. The bias types include:

[0241] If clean bedding and mite removal are pushed (for dust mite allergy), but the feedback data shows that cleaning does not relieve coughing (actually for pollen allergy), mark the intervention measure as mismatched with the allergen.

[0242] If the PEF measurement is increased to 3 times / day, and the feedback after execution captures the PEF decrease in time to avoid attacks, mark the intervention as effective and no adjustment is needed.

[0243] Low priority bias: including data record delay, symptom option omission, not involving model optimization adjustment.

[0244] S9: Based on the bias between the risk assessment results and the feedback data, optimize the dynamic weight feature extraction model, the logistic regression model and the XGBoost model parameters, and re-execute steps S2-S6.

[0245] Optimize the dynamic weight feature extraction model for the overestimation of the AQI exceeding the standard for a long time weight:

[0246] Retrieve feedback samples (≥500) in the past 3 months, calculate the exacerbation rate according to the AQI grouping, and based on the actual difference / expected difference, reduce the weight from 0.3 to 0.09. The optimization goal is that the actual impact of the AQI exceeding the standard for a long time feature and the new weight bias ≤2%.

[0247] For unreasonable medication compliance adjustment coefficient, bias: high compliance original coefficient 0.7, actual need 0.6:

[0248] The PEF variability impact degree reduction rate = (1 - adjustment coefficient) x 100%. The high compliance (C ≥ 0.8) coefficient is corrected from 0.7 to 0.6.

[0249] Extended to the full interval: medium compliance (0.5 ≤ C < 0.8) coefficient 0.8, low compliance (C < 0.5) coefficient 1.1. Optimization goal: the adjusted PEF variability impact degree and feedback data matching degree ≥ 95%.

[0250] Logistic regression model optimization:

[0251] For low risk underestimation, bias: 15 cases of exacerbation in 100 low risk samples, underestimation bias rate 15%. Lower the low risk threshold from <20% to <15%. Increase the PEF variability feature coefficient: original β2 = 0.5, up to 1.56. Optimization goal: low risk underestimation bias rate ≤ 5%, model test set AUC ≥ 0.88.

[0252] For high risk overestimation, assume 10 cases of remission in 100 high risk samples, overestimation bias rate 10%

[0253] Test different thresholds, increase the high risk threshold from ≥ 50% to ≥ 55%. Reduce the allergen exposure intensity coefficient: original β4 = 0.3, down to 0.12. Optimization goal: high risk overestimation bias rate ≤ 3%, high risk sample actual attack rate ≥ 88%.

[0254] XGBoost model optimization:

[0255] For intermediate risk bias rate 20%, assume 20 actual low / high risk in 100 intermediate risk samples. Tree depth from 5 to 6, learning rate from 0.05 to 0.03, tree number from 200 to 250. Subdivide the intermediate risk interval: medium-low risk (20%-35%) focuses on historical frequency of attack, medium-high risk (35%-50%) focuses on PEF variability. Optimization goal: intermediate risk classification accuracy rate ≥ 92%, bias rate ≤ 8%.

[0256] A child asthma acute attack risk assessment system comprises: a multi-modal data acquisition and preprocessing module for acquiring static basic data, dynamic physiological data, environmental interaction data and subjective symptom data of a target child, and preprocessing the multi-modal data. A dynamic weight feature extraction model is used to extract specified features to construct a basic feature vector, adjust dynamic feature weights based on individual differences of the child and real-time scenarios, construct a weighted feature vector based on the basic features and dynamic feature weights, and construct a fusion feature vector by incorporating the child's recent medication compliance. A logistic regression model is used to input the fusion feature vector and output the probability of asthma acute attack occurring in a specified future time period. When the probability is less than a first probability threshold, a routine care plan instruction is sent to a scheme pushing module; when the probability is greater than a second probability threshold, an emergency intervention scheme instruction is sent to the scheme pushing module; otherwise, an evaluation instruction is sent to a risk level assessment module. An XGBoost model is used to input the fusion feature vector into the XGBoost model after receiving the evaluation instruction, and output a preliminary risk level, which includes low risk, medium risk and high risk. A risk level calibration module is used to calculate the cosine similarity between the current fusion feature vector and the historical attack feature vector, calibrate the preliminary risk level based on the cosine similarity, and obtain a calibrated risk level. A dynamic intervention scheme generation and pushing module is used to generate a dynamic intervention scheme based on the calibrated risk level. When the calibrated risk level is low risk, a corresponding routine care plan is generated; when the risk level is medium risk, an intensive monitoring scheme is generated; and when the risk level is high risk, a corresponding emergency intervention scheme is output.

Claims

1. A method of assessing the risk of an acute asthma exacerbation in a child, characterized in that, The method comprises the following steps: S1: acquiring multi-modal data of a target child including static basic data, dynamic physiological data, environmental interaction data and subjective symptom data, and preprocessing the multi-modal data; S2: constructing a dynamic weight feature extraction model to extract specified features and construct a basic feature vector, acquiring dynamic feature weights based on individual differences of the child and real-time scenes, constructing a weighted feature vector based on the basic feature vector and the dynamic feature weights, and fusing the child's recent medication compliance to construct a fusion feature vector; S3: inputting the fusion feature vector into a logistic regression model to output a probability of asthma acute attack occurring in a future specified period of time, pushing a routine care plan when the probability is less than a first probability threshold, pushing an emergency intervention plan when the probability is greater than a second probability threshold, and otherwise executing step S4; S4: inputting the fusion feature vector into an XGBoost model to output a preliminary risk level, the preliminary risk level including low risk, medium risk and high risk; S5: calculating a cosine similarity between the current fusion feature vector and a historical attack feature vector, calibrating the preliminary risk level based on the cosine similarity to obtain a calibrated risk level; S6: generating a dynamic intervention plan based on the calibrated risk level, generating a corresponding routine care plan when the calibrated risk level is low risk, generating an intensive monitoring plan when the calibrated risk level is medium risk, and outputting a corresponding emergency intervention plan when the calibrated risk level is high risk; The specific process of step S2 is as follows: S21: extracting specified basic features through a basic feature layer, including the target child's acute attack frequency in a specified period of time, PEF diurnal variability, AQI exceeding cumulative duration, and daily average exposure intensity of allergen, to obtain a basic feature vector; Acute attack frequency F1: counting the total number of asthma acute attacks of the target child in a specified period of time; PEF diurnal variability F2: (daytime maximum PEF value-nighttime minimum PEF value) ÷ daytime maximum PEF value × 100%; AQI exceeding cumulative duration F3: counting the duration of AQI > 150 on the current day; Daily average exposure intensity of allergen F4: for the allergen to which the child is allergic, counting the sum of allergen concentration × contact duration in each period on the current day, and then dividing by 24 hours to obtain the daily average intensity; The basic feature vector is F = [F1, F2, F3, F4]; S22: the dynamic weight layer gives different specified dynamic feature weights through an attention mechanism, multiplies the basic features by the corresponding dynamic feature weights to obtain a dynamic weight vector: S23: multiplying the basic feature vector by the dynamic weight vector to generate a weighted feature vector; the calculation formula is: weighted feature vector = basic feature vector × dynamic weight vector; S24: incorporating the medication compliance of the child in a specified period of time, the number of days of taking medication on time ÷ the total number of days of taking medication, denoted as C, is a risk adjustment factor, with a value range of 0-1, which is used to adjust the value of the weighted feature vector to reflect the risk mitigation effect of regular medication or the risk amplification effect of irregular medication, and to construct a fusion feature vector: Calculate the medication compliance C: Retrieve the child's medication records in the past 14 days and calculate according to the formula: C = number of days of taking medication on time ÷ total number of days of taking medication; Set the medication adherence adjustment rule: High adherence, i.e. C≥0.8: multiply the three types of real-time risk features of PEF variation rate, AQI exceeding duration, and allergen exposure intensity in the weighted feature vector by the adjustment coefficient 0.7; Medium adherence, i.e. 0.5≤C<0.8: the adjustment coefficient is 0.9; Low adherence, i.e. C<0.5: the adjustment coefficient is 1.2; Generate the fusion feature vector.

2. The method of claim 1, wherein the method further comprises: The specific process of preprocessing the multi-modal data in step S1 is as follows: S11: preset the verification rules of each modality data, and verify the corresponding modality data based on the verification rules of each modality data. The implicit abnormal data that cannot be covered by the verification rule is detected by the isolation forest algorithm for unsupervised anomaly detection; S12: perform missing value completion processing on the calibrated multi-modal data: For dynamic physiological data, use an LSTM-based time series prediction model for missing value completion. For subjective symptom data, use historical same period data and daily environmental data for correlation completion. For environmental interaction data, use adjacent node data interpolation completion. For static basic data, use prompt completion + default rule for completion; S13: standardize the completed multi-modal data.

3. A method of assessing the risk of an acute asthma exacerbation in a child according to claim 2, wherein, The dynamic weight feature extraction model in step S2 includes a basic feature layer, a dynamic weight layer, and a fusion feature layer. The specific process of extracting specified features is as follows: S21: extract specified basic features through the basic feature layer, including the acute attack frequency of the target child in a specified time period, the PEF diurnal variation rate, the cumulative duration of AQI exceeding, and the daily average exposure intensity of allergens, to obtain a basic feature vector; S22: the dynamic weight layer gives different specified basic features dynamic differential weights through an attention mechanism, multiplies the basic features by the corresponding dynamic differential weights to obtain a dynamic weight vector: S23: multiply the basic feature vector and the dynamic weight vector to generate a weighted feature vector; S24: incorporate the medication adherence of the child in the specified time period as a risk adjustment factor. By adjusting the numerical value of the weighted feature vector, the regular medication can reflect the risk mitigation effect or the irregular medication can amplify the risk. A fusion feature vector is constructed.

4. The method of claim 3, wherein the method further comprises: The specific process of inputting the fusion feature vector into the logistic regression model in step S3 to output the probability of asthma acute attack in the future specified time period is as follows: S31: load the pre-trained model parameters, including feature coefficients β0, β1, β2, β3, and β4, where β0 is the intercept term, and β1-β4 correspond to the four dimensions of the fusion feature vector; S32: input the fusion feature vector into the logistic regression model for linear combination calculation to obtain a linear prediction value z reflecting the risk level: S33: convert the linear value to a probability through Sigmoid function mapping.

5. The method of claim 1, wherein the method further comprises: The specific process of step S4 is as follows: S41: load the XGBoost model parameters, including tree structure parameters, feature attention coefficients, and basic probability mapping parameters; S42: based on the loaded feature attention coefficients, assign an initial importance weight to each dimension of the fusion feature vector; S43: The XGBoost model calculates through parallel splitting of multiple decision trees, integrates the results, and outputs the original risk score; S44: The original risk score is converted into the probability of onset through an improved Sigmoid function; S45: The probability is mapped into a preliminary risk level.

6. The method of claim 4, wherein the method further comprises: The specific process of step S5 is as follows: S51: Acquire the records of acute attacks of the target child in the past year, and construct the historical attack feature vector of the target child; S52: Calculate the cosine similarity between the current fusion feature vector and H1, H2, and H3, respectively; S53: Take the average of the multiple historical attack similarities as the final calibration basis; S54: Calibrate the preliminary risk level based on the average similarity and the preset calibration rule.

7. The method of claim 1, wherein the method further comprises: It also includes setting a feedback adjustment mechanism, and the specific process is as follows: S7: Acquire the feedback data of the target child, including intervention execution data and disease change data; S8: Establish a mapping relationship between the feedback data and the risk assessment results, evaluate the accuracy of the risk assessment results according to the mapping relationship, and quantify the deviation between the risk assessment results and the feedback data; S9: Based on the deviation between the risk assessment results and the feedback data, optimize the dynamic weight feature extraction model, the logistic regression model and the XGBoost model, and re-execute steps S2-S6.

8. A system for assessing the risk of an acute asthma exacerbation in a child for implementing a method for assessing the risk of an acute asthma exacerbation in a child according to any one of claims 1 to 7, characterized in that, It includes: A multi-modal data acquisition and preprocessing module is used to acquire the target child's static basic data, dynamic physiological data, environmental interaction data, and subjective symptom data, and to preprocess these multi-modal data; A dynamic weight feature extraction model is used to extract specified features to construct a basic feature vector, adjust the dynamic feature weight based on individual differences and real-time scenarios, construct a weighted feature vector based on the basic feature and the dynamic feature weight, and construct a fusion feature vector by incorporating the child's recent medication compliance; A logistic regression model is used to input the fusion feature vector, output the probability of acute asthma attack in a specified future time period, send a push regular care plan instruction to the scheme pushing module when the probability is less than the first probability threshold, send a push emergency intervention scheme instruction to the scheme pushing module when the probability is greater than the second probability threshold, and otherwise send an evaluation instruction to the risk level assessment module; An XGBoost model is used to input the fusion feature vector into the XGBoost model after receiving the evaluation instruction, output a preliminary risk level, and the preliminary risk level includes low risk, medium risk and high risk; A risk level calibration module is used to calculate the cosine similarity between the current fusion feature vector and the historical attack feature vector, calibrate the preliminary risk level based on the cosine similarity, and obtain the calibrated risk level; A dynamic intervention scheme generation and pushing module is used to generate a dynamic intervention scheme based on the calibrated risk level, generate a corresponding regular care plan when the calibrated risk level is low, generate an intensive monitoring scheme when the calibrated risk level is medium, and output a corresponding emergency intervention scheme when the calibrated risk level is high.

Images