Early warning method and system for acute exacerbation risk of chronic obstructive pulmonary disease based on home monitoring data and environmental factors

Abstract

The invention discloses a chronic obstructive pulmonary disease acute exacerbation risk early warning method and system based on home monitoring data and environmental factors, and belongs to the technical field of medical health monitoring and risk early warning. The method comprises the following steps: acquiring multi-dimensional data (symptom score, FEV1 trend, night blood oxygen saturation, medication compliance and weather data) of a user; pre-processing the multi-dimensional data and inputting the pre-processed multi-dimensional data into the trained machine learning prediction model; the model outputs an acute exacerbation risk probability value in the future 7 days; when the risk probability exceeds a preset threshold value, graded early warning information and personalized intervention suggestions are sent. The system comprises a data acquisition module, a data preprocessing module, a model prediction module, an early warning pushing module and a terminal interaction module. Multi-source data are integrated, through targeted feature extraction and model training, early-stage accurate early warning of chronic obstructive pulmonary acute exacerbation is achieved, meanwhile, personalized intervention suggestions are provided, early warning-intervention-feedback closed-loop management is supported, operation is convenient, cost is controllable, the acute exacerbation occurrence rate and the hospitalization rate of patients can be effectively reduced, and the patient experience is improved. The method is suitable for popularization and application in family and basic medical scenes.

Description

Technical Field

[0001] This invention relates to the field of medical and health monitoring and risk warning technology, specifically to a method and system for early warning of acute exacerbations of COPD based on home monitoring data and environmental factors. Background Technology

[0002] Chronic obstructive pulmonary disease (COPD) is a common, preventable, and treatable chronic airway disease characterized by persistent respiratory symptoms and airflow limitation that is not fully reversible and progresses. An acute exacerbation of COPD refers to an acute worsening of a patient's respiratory symptoms, requiring adjustments to their regular medication regimen. It is a significant cause of disease progression, decreased quality of life, and increased hospitalization and mortality rates. Clinical data shows that COPD patients experience an average of 2-3 acute exacerbations per year, each causing irreversible damage to lung function and severely impacting patient prognosis.

[0003] Currently, early warning of COPD exacerbations mainly relies on patients' regular outpatient follow-up and self-reported symptoms. Outpatient follow-up is usually spaced 1-3 months apart, making real-time monitoring and early warning difficult; while patients' self-reported symptoms are often delayed, with most patients experiencing an acute exacerbation only after significant cough, sputum production, and worsening dyspnea have occurred, missing the optimal intervention window. Furthermore, existing early warning technologies often focus only on single-dimensional data (such as monitoring only lung function or only symptom scores), ignoring the impact of environmental factors (such as temperature, humidity, and AQI) and medication adherence on acute exacerbations, resulting in low accuracy and failing to meet actual clinical needs.

[0004] In existing technologies, some patents attempt to combine multi-source data for early warning of acute exacerbations of COPD, but they have the following drawbacks: First, the data fusion method is simple and does not extract specific features for different types of data (time series data, numerical data); second, the predictive model has insufficient generalization ability and lacks large-scale clinical sample training and validation; third, the early warning and intervention are not closely integrated, and can only output the risk level, but cannot provide personalized intervention suggestions, making it difficult to effectively guide clinical practice and patient self-management.

[0005] Therefore, there is an urgent need for a method and system that can integrate multi-source home monitoring data and environmental factors to achieve early and accurate warning of COPD acute exacerbations and provide personalized intervention suggestions simultaneously, in order to solve the technical problems of delayed warning, low accuracy and insufficient intervention in existing technologies. Summary of the Invention

[0006] 3.1 Purpose of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies, such as delayed early warning of COPD exacerbations, low accuracy, focus on only single-dimensional data, and lack of personalized intervention suggestions. This invention provides a method and system for early warning of COPD exacerbation risk based on home monitoring data and environmental factors. This enables early and accurate early warning of COPD exacerbations and provides targeted intervention suggestions for patients and medical staff, thereby reducing the incidence of exacerbations and hospitalization rates and improving patients' quality of life.

[0008] To achieve the above objectives, this invention provides a method for early warning of COPD acute exacerbation risk based on home monitoring data and environmental factors, comprising the following steps:

[0009] Data acquisition steps: Acquire multidimensional data from users, including daily self-reported symptom scores (mMRC score, CAT score), FEV1 trend data measured by a home spirometer, nighttime blood oxygen saturation data monitored by a smart oxygen meter (minimum value, average value, duration below 90%), medication adherence data recorded by a smart pillbox (medication time, dosage, number of missed doses, and make-up doses), and local weather data (temperature, humidity, AQI) obtained through a public API interface. Data acquisition adopts a combination of real-time collection and periodic uploading. Physiological monitoring data and medication data are uploaded every 24 hours, while weather data is updated synchronously in real time.

[0010] Model input steps: Preprocess the acquired multidimensional data, specifically including outlier removal (3σ principle), missing value imputation (linear interpolation / K-nearest neighbor algorithm), and standardization (min-max standardization to the [0,1] interval); input the preprocessed standardized data into the trained machine learning prediction model (LSTM network or gradient boosting tree model).

[0011] Risk output steps: The machine learning prediction model extracts features from the input data, where the LSTM network is used to extract time-series features (such as FEV1 trend and blood oxygen saturation fluctuations), and the gradient boosting tree is used to extract non-linear correlations between features; the model outputs the risk probability value (0-100%) of the user's acute exacerbation of COPD in the next 7 days through feature fusion and analysis.

[0012] Early warning intervention steps: Preset three-level risk thresholds (low risk ≤30%, medium risk 30%-70%, high risk >70%), compare the risk probability value with the preset thresholds; when the risk probability value exceeds the threshold, send graded early warning information (low risk prompt, medium risk reminder, high risk alarm) to the patient terminal and medical staff terminal, and generate personalized intervention suggestions (medication adjustment, lifestyle intervention, medical visit reminder) based on the patient's historical data, current symptoms and environmental factors.

[0013] This invention also provides a COPD acute exacerbation risk early warning system for implementing the above method, comprising a data acquisition module, a data preprocessing module, a model prediction module, an early warning push module, and a terminal interaction module, the functions of each module being as follows:

[0014] Data acquisition module: Composed of home monitoring devices (home spirometer, smart oxygen meter, smart pillbox), user terminal (mobile APP) and API interface, it is used to collect patients' physiological monitoring data and medication data in real time, simultaneously acquire environmental weather data, and upload the data to the cloud server.

[0015] Data preprocessing module: Deployed on a cloud server, it is used to clean, remove outliers, fill in missing values, and standardize the collected multidimensional data to generate standardized feature data suitable for model input, ensuring data quality and model prediction accuracy.

[0016] Model prediction module: It has a built-in LSTM network or gradient boosting tree prediction model trained on a large number of clinical samples. It receives preprocessed feature data, and outputs the probability value of acute exacerbation risk in the next 7 days through feature extraction, fusion and analysis. It supports dynamic model updates and continuously optimizes prediction performance based on new clinical data and feedback.

[0017] Early warning push module: It is used to preset three-level risk thresholds, trigger corresponding level early warnings based on the risk probability values ​​output by the model, push early warning information to patients and medical staff through multiple channels (APP, SMS, telephone, WeChat), and generate personalized intervention suggestions based on individual patient conditions.

[0018] Terminal interaction module: including patient APP and medical staff management platform. Patients can view risk level, early warning information, intervention suggestions and report changes in symptoms through the APP; medical staff can view patient monitoring data, risk trends and intervention feedback through the management platform, and issue personalized intervention instructions to achieve closed-loop management.

[0019] Compared with the prior art, the present invention has the following advantages:

[0020] Comprehensive data dimensions for more accurate early warning: This invention integrates patient home physiological monitoring data, medication adherence data, and environmental factor data, breaking through the limitations of existing technologies with a single data dimension; through targeted data preprocessing and feature extraction, it fully explores the correlation between various data dimensions, significantly improving the accuracy of predicting the risk of acute exacerbation (verified accuracy ≥85%, recall ≥80%), and achieving early and accurate early warning.

[0021] The invention combines early warning and intervention closely, making it highly practical: It can not only output risk levels, but also generate personalized intervention suggestions based on individual patient conditions (historical data, current symptoms, environmental factors). At the same time, it supports real-time intervention by medical staff, realizing closed-loop management of "early warning-intervention-feedback", effectively guiding patients' self-management and clinical intervention, and reducing the incidence of acute exacerbations.

[0022] Easy to operate and promote: This invention realizes data collection and early warning push based on home monitoring devices and smart terminals, eliminating the need for patients to frequently visit the hospital, and is easy to operate; the system has strong compatibility, can be connected to existing home monitoring devices and hospital information systems, has controllable costs, and is easy to promote and apply in primary medical institutions and home scenarios.

[0023] Strong model generalization ability: This invention uses a large-scale clinical sample (≥1000 cases) to train the prediction model. Through cross-validation and hyperparameter optimization, the generalization ability of the model is guaranteed, and it can be applied to COPD patients of different ages and different disease severity.

[0024] Figure 1 This is a diagram illustrating the overall architecture of the COPD acute exacerbation risk warning system of the present invention.

[0025] Figure 2 This is a flowchart of the method for early warning of acute exacerbation of COPD according to the present invention;

[0026] Figure 3 This is a flowchart illustrating the training process of the machine learning prediction model of this invention.

[0027] Figure 4 This is a closed-loop logic diagram of the early warning push and terminal interaction of the present invention.

[0028] The deployment of the early warning system described in this invention comprises four parts: a cloud server, home monitoring devices, patient terminals, and medical staff terminals.

[0029] Cloud server: Alibaba Cloud ECS server, configured with 8 cores and 16GB of memory, running Windows Server 2019 operating system, with MySQL database (for data storage), Python 3.8 environment (for model training and prediction) and Flask framework (for interface development) installed; data storage uses AES encryption algorithm to ensure patient privacy and security.

[0030] Home monitoring equipment: A home spirometer (measurement range 0-10L, accuracy ±5%), a smart oxygen meter (measurement range 70%-100%SpO2, accuracy ±2%), and a smart pillbox (supports medication time reminders, missed dose records, and Bluetooth 4.0 communication) certified by the National Medical Products Administration are selected.

[0031] Patient terminal: Supports smartphones with Android 8.0 and above and iOS 12.0 and above, with a dedicated APP installed (supporting data reporting, early warning reception, and symptom feedback).

[0032] Healthcare worker terminals: computers (Windows / macOS systems) and smartphones, with a healthcare management platform deployed (supporting patient data viewing, risk monitoring, and intervention command issuance).

[0033] This embodiment uses an LSTM network as the prediction model, and the specific training process is as follows:

[0034] Sample data collection: Home monitoring data and clinical follow-up records of 1200 COPD patients (disease duration 3-10 years, age 45-75 years) in the Department of Respiratory Medicine of a tertiary hospital were collected for 6 consecutive months. Among them, 420 patients had acute exacerbations and 780 patients did not have acute exacerbations. The monitoring data included daily mMRC score (0-4), CAT score (0-40), FEV1 value (L), nocturnal blood oxygen saturation data, medication adherence data, and local temperature (°C), humidity (%), and AQI value on the corresponding dates.

[0035] Sample labeling and division: Each patient was labeled based on whether an acute exacerbation occurred within 7 days after a certain date (diagnosed by clinicians based on symptoms such as increased cough, sputum production, and dyspnea, as well as pulmonary function test results). An acute exacerbation was recorded as 1, and no exacerbation was recorded as 0. The 1200 samples were divided into a training set (840 cases) and a test set (360 cases) in a 7:3 ratio.

[0036] Data preprocessing: Outliers in numerical data such as FEV1 and blood oxygen saturation were removed using the 3σ principle (removal rate of approximately 3.2%); missing values ​​were filled using linear interpolation (missing rate of approximately 2.5%); and all data were normalized to the [0,1] interval using min-max standardization.

[0037] LSTM Model Construction and Training: A 3-layer LSTM network was constructed, with an input layer dimension of 10 (corresponding to 10 feature dimensions: mMRC score, CAT score, FEV1 value, minimum nighttime blood oxygen saturation, average nighttime blood oxygen saturation, duration of nighttime blood oxygen saturation below 90%, medication adherence rate, temperature, humidity, and AQI), 64 neurons in the hidden layer, and 1 neuron in the output layer (outputting the risk probability value). The cross-entropy loss function was used as the optimization objective, with the Adam optimizer (learning rate 0.001, decay coefficient 0.9), 100 iterations, and a batch size of 32.

[0038] Model validation: The test set was input into the trained LSTM model, and the model performance metrics were calculated: accuracy 86.7%, recall 82.1%, and F1 score 84.3%, which meets the clinical early warning requirements.

[0039] Taking a 65-year-old male patient with COPD (5-year disease duration, with 2 acute exacerbations per year in the past) as an example, the implementation process of the method described in this invention is as follows:

[0040] Data collection: The patient's FEV1 value was measured daily using a home spirometer (the measured value on the day was 2.1L, and the trend over the past 7 days was downward). Nighttime blood oxygen saturation was monitored using a smart oxygen meter (the lowest value was 88%, and the duration of the value below 90% was 45 minutes). The mMRC score (level 2) and CAT score (22 points) were reported via the APP. The smart pillbox recorded one missed dose of long-acting bronchodilator on the day. The system also simultaneously obtained the local weather data for the day (temperature 12℃, humidity 45%, AQI 180, moderate pollution).

[0041] Data preprocessing: The system cleans (removes outliers), fills (removes missing values), and standardizes the above data to generate standardized feature vectors.

[0042] Risk prediction: The standardized feature vector is input into the LSTM prediction model, and the model outputs a probability value of 78% for acute exacerbation within the next 7 days.

[0043] Warning push: When the risk probability value (78%) exceeds the high risk threshold (70%), the system immediately sends a high risk alert to the patient's APP (red pop-up notification), and at the same time sends an SMS reminder and a telephone voice reminder to the patient's mobile phone; and sends high risk warning information (including patient basic information, current monitoring data, and risk probability value) to the patient's attending physician's medical management platform.

[0044] Personalized Intervention: The system generates personalized intervention recommendations: ① Medication Adjustment: Temporarily increase the frequency of short-acting bronchodilator inhalation (3 times daily, 1 inhalation each time); ② Lifestyle Intervention: Avoid going out (moderate AQI pollution), keep warm, and avoid catching a cold; ③ Medical Visit Reminder: It is recommended to visit the outpatient clinic for a follow-up visit within 24-48 hours. After reviewing the patient's data through the medical management platform, the attending physician issues a confirmation intervention instruction and schedules the patient's follow-up outpatient visit for the next day.

[0045] Feedback and Follow-up: Patients confirm receipt of warning information and intervention suggestions through the APP and report that their current cough symptoms have worsened; the next day, patients go to the outpatient clinic for a follow-up visit, and the doctor adjusts the medication plan according to the condition. The system continuously monitors the patient's subsequent data and tracks the intervention effect.

Claims

1. A method for predicting the risk of acute exacerbation of chronic obstructive pulmonary disease, characterized by, Comprise the following steps: (1) Data acquisition step: acquire the user's multidimensional data, the multidimensional data includes daily self-reported symptom score, family spirometer measured forced expiratory volume in one second (FEV1) trend data, intelligent finger oxygen instrument monitored night blood oxygen saturation data, smart medicine box recorded medication compliance data, and local weather data obtained through public API interface; wherein, the symptom score includes modified British Medical Research Council dyspnea scale (mMRC) score and chronic obstructive pulmonary disease assessment test (CAT) score; the weather data includes temperature, humidity and air quality index (AQI); (2) Model input step: after preprocessing the multidimensional data, input into the trained machine learning prediction model; the machine learning prediction model is long short-term memory network (LSTM) or gradient boosting tree model; (3) Risk output step: the machine learning prediction model extracts and analyzes the input data, and outputs the risk probability value of the user's acute exacerbation of chronic obstructive pulmonary disease within 7 days in the future; (4) Early warning intervention step: preset risk threshold, compare the risk probability value with the preset risk threshold, when the risk probability value exceeds the preset risk threshold, send hierarchical warning information to the patient and designated medical staff, and generate individualized intervention suggestions.

2. A system for early warning of risk of acute exacerbation of chronic obstructive pulmonary disease, characterized in that, Comprise: (1) Data acquisition module: used for collecting the user's multidimensional data, the multidimensional data includes daily self-reported symptom score, family spirometer measured FEV1 trend data, intelligent finger oxygen instrument monitored night blood oxygen saturation data, smart medicine box recorded medication compliance data, and local weather data obtained through public API interface; (2) Data preprocessing module: used for cleaning, outlier rejection, missing value filling and standardization processing of the multidimensional data obtained by the data acquisition module; (3) Model prediction module: built-in trained machine learning prediction model, used for receiving the standardized data output by the data preprocessing module, extracting and analyzing features, and outputting the risk probability value of the user's acute exacerbation of chronic obstructive pulmonary disease within 7 days in the future; (4) Early warning push module: used for presetting risk threshold, comparing the risk probability value output by the model prediction module with the preset risk threshold, when the risk probability value exceeds the preset threshold, sending hierarchical warning information and individualized intervention suggestions to the patient terminal and medical staff terminal; (5) Terminal interaction module: including patient terminal and medical staff terminal, used for receiving warning information and intervention suggestions, while supporting patient feedback symptom change information and medical staff issued intervention instructions.

3. The method of claim 1, wherein, In the data acquisition step, the medication compliance data includes medication time, medication dose, missed dose frequency and make-up situation; the night blood oxygen saturation data includes the minimum value, average value and duration of less than 90% of the night blood oxygen saturation.

4. The method of claim 1, wherein, The data preprocessing step specifically includes: removing outliers in multidimensional data using the 3σ principle; filling in missing values using linear interpolation or K-nearest neighbor algorithm; normalizing the processed data to the [0, 1] interval using the min-max standardization method, and the standardization formula is: x'=(x-x_min) / (x_max-x_min), where x is the original data, x_min is the minimum value of the dimension data, x_max is the maximum value of the dimension data, and x' is the standardized data.

5. The method of claim 1, wherein, The training process of the machine learning prediction model includes: (1) Collecting sample data: collecting historical multidimensional monitoring data of at least 1000 cases of patients with chronic obstructive pulmonary disease and corresponding diagnosis records of acute exacerbation, which are confirmed by clinicians according to symptoms and examination results; (2) Sample labeling and division: taking whether the sample data will occur acute exacerbation within the next 7 days as the label, and recording 1 for occurrence and 0 for non-occurrence; dividing the labeled sample data into training set and test set according to the ratio of 7:3; (3) Model training: inputting the training set into the initial machine learning model, taking cross-entropy loss function as the optimization objective, using Adam optimizer to adjust the model hyperparameters, and iteratively training until the loss function converges; (4) Model verification: inputting the test set into the trained model, calculating the accuracy, recall rate and F1 score of the model, and when the accuracy is ≥85%, the recall rate is ≥80%, the model training is completed.

6. The method of claim 1, wherein, The preset risk threshold includes low risk threshold, medium risk threshold and high risk threshold, wherein the low risk threshold is ≤30%, the medium risk threshold is 30%-70%, and the high risk threshold is >70%; and the corresponding graded warning information includes low risk prompt, medium risk reminder and high risk alarm.

7. The method of claim 1, wherein, The individualized intervention suggestions include drug adjustment suggestions, lifestyle intervention suggestions and medical advice; wherein the drug adjustment suggestions are preset by medical staff according to the patient's historical medication data and current risk level, and the lifestyle intervention suggestions include keeping warm, avoiding outdoor exposure, quitting smoking and appropriate exercise.

8. The system of claim 2, wherein, The data acquisition module further includes a data synchronization unit for synchronizing the collected multidimensional data to the cloud server in real time, and the cloud server uses AES encryption algorithm to encrypt and store the data, ensuring data security.

9. The system of claim 2, wherein, The warning pushing module supports multi-channel pushing, including APP pushing, SMS reminder, telephone voice reminder and WeChat public number pushing; wherein the high risk alarm adopts the combination of APP pushing+SMS reminder+telephone voice reminder, the medium risk reminder adopts the combination of APP pushing+SMS reminder, and the low risk prompt adopts the APP pushing mode.

10. The system of claim 2, wherein, The terminal interaction module also supports medical staff to view patient historical monitoring data, risk prediction trend and intervention feedback record through the terminal, realizing the closed-loop management of warning-intervention-feedback.

Images