A Depression Monitoring and Intervention System Based on Internet Hospital and Heart Rate Variability
The depression monitoring and intervention system, which combines multimodal physiological signal acquisition and multilayer perceptron neural network with latent class hybrid model, solves the problems of early identification difficulties and missed diagnoses by traditional scales, and achieves efficient identification and personalized intervention of depression, especially improving the accuracy in people with comorbid anxiety and depression.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- RENJI HOSPITAL AFFILIATED TO SHANGHAI JIAO TONG UNIV SCHOOL OF MEDICINE
- Filing Date
- 2026-01-08
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies for monitoring and intervening in depressive states have problems such as difficulty in early identification, high rate of missed diagnoses with traditional scales, and insufficient resources of psychiatrists, and are particularly ineffective in large-scale screening.
A multimodal physiological signal acquisition module was used to simultaneously collect heart rate variability, skin conductance fluctuation rate, and respiratory rate data. The signal processing module performed preprocessing and feature extraction, and a multilayer perceptron neural network was used to infer depression levels. The results were combined with a latent category mixture model for user segmentation and personalized intervention.
It enables early identification and personalized intervention for depressive states, improves identification accuracy, and reduces medical resource consumption. In particular, it improves identification accuracy by 12%–18% in people with comorbid anxiety and depression, and shortens the recovery time of HRV indicators through personalized intervention.
Smart Images
Figure CN122074982A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of depression monitoring and intervention technology, and in particular to a depression monitoring and intervention system based on internet hospitals and heart rate variability. Background Technology
[0002] Depressive disorders are the third leading cause of disease burden globally (WHO 2023 report), and their early identification faces significant challenges: More than 60% of patients initially present with only autonomic dysfunction, lacking typical emotional symptoms; traditional scales such as PHQ-9 / HAMD rely on subjective statements, resulting in a 45% missed diagnosis rate in community screening; and the proportion of psychiatrists in the population is less than 0.1 per 100,000, which cannot support large-scale screening.
[0003] A completely new system for monitoring and intervening in depressive states based on heart rate variability is needed. Summary of the Invention
[0004] The main objective of this invention is to solve the technical problem of the failure of existing methods for monitoring and intervening in depressive states. A system for monitoring and intervening in depressive states based on internet hospitals and heart rate variability includes: A multimodal physiological signal acquisition module is used to simultaneously acquire raw heart rate variability signals, skin conductance fluctuations, and respiratory rate data. The signal processing module is used to preprocess the original signal and extract features to obtain multi-dimensional feature vectors; The depression level inference module constructs a multilayer perceptron neural network. The input is a multidimensional feature vector containing multidimensional features of heart rate variability, skin conductance fluctuation rate, and respiratory rate statistics. The output is the probability value of no depression, mild depression, and moderate to severe depression level.
[0005] As a preferred technical solution, preprocessing of the original signal includes: The raw heart rate variability signal was decomposed into a 5-level wavelet based on the Sym8 wavelet basis. After decomposition, the high-frequency coefficients of the first to third levels were processed by hard thresholding to remove electromyographic interference, and the low-frequency coefficients of the fourth to fifth levels were processed by soft thresholding to eliminate baseline drift. The processed signal was then reconstructed by wavelet to restore the denoised pulse wave signal. For the RR interval sequence, linear interpolation is used to resample it to a 1Hz isochronous sequence, and then a 5-point moving average filter is used to remove instantaneous noise, finally obtaining stable HRV time series data. As a preferred technical solution, the feature extraction includes: Based on the preprocessed HRV time series data, multi-dimensional features are calculated through the following steps: In the time domain features, the standard deviation of the normal RR interval (SDNN) is obtained by calculating the standard deviation of all RR intervals within a 5-minute window, and the root mean square (RMSSD) of the difference between adjacent RR intervals is obtained by taking the square root of the average of the squares of the differences between adjacent RR intervals. In the frequency domain features, the power spectrum of the resampled HRV sequence is estimated by fast Fourier transform. The low-frequency power LF is the power integral of the 0.04-0.15 Hz band, and the high-frequency power HF is the power integral of the 0.15-0.4 Hz band. The LF / HF ratio is the ratio of the power values of the two. In the nonlinear features, the approximate entropy ApEn is calculated using parameters with an embedding dimension of 2 and a similarity tolerance of 0.2 times the standard deviation. The sample entropy SampEn uses the same embedding dimension and tolerance parameters. The short axis SD1 and long axis SD2 of the Poincaré plot are obtained by calculating the scatter distribution characteristics of the RR interval sequence. The skin conductance response variability was obtained by calculating the ratio of the standard deviation to the mean of the EDA data within a 1-minute window; The respiratory rate is characterized by the mean and standard deviation within a 5-minute window. As a preferred technical solution, the depression level inference module uses a multilayer perceptron neural network comprising an input layer, three hidden layers, and an output layer. The three hidden layers have 64, 32, and 16 neurons respectively, and the activation function is ReLU. The output layer includes three neurons, corresponding to no depression, mild depression, and moderate to severe depression, and the activation function is Softmax.
[0006] As a preferred technical solution, an intervention system based on a depression state monitoring system is configured as follows: By using a latent category mixture model to analyze the dynamic trajectory of users' HRV over the past 7 days, including data from resting state and daily activity state, users are divided into typical declining users whose LF / HF ratio decreases by ≥30% during stress and atypical rising users whose LF / HF ratio increases by ≥20% during stress. Based on user segmentation, intervention is carried out on users.
[0007] As a preferred technical solution, the intervention system further includes a wearable device and a personalized intervention module, wherein the personalized intervention module is configured as follows: For typical declining users, when a depression warning is triggered, a course is pushed to them. The course uses an inhalation-breath-exhalation rhythm and is guided synchronously through screen animations and voice prompts. During the course, the RMSSD value of HRV is monitored in real time. The training ends when the RMSSD value rises back to more than 90% of the baseline level.
[0008] As a preferred technical solution, the personalized intervention module is configured as follows: For users with atypical elevated muscle type, after triggering the warning, progressive muscle relaxation training is initiated. Users are guided to alternately tighten and relax their muscles in the order of hand-forearm-upper arm-shoulder-face-trunk-lower limb. During the training, the skin conductance response value is monitored. The training ends when the skin conductance response value drops to less than 80% of the pre-warning level. As a preferred technical solution, the personalized intervention module is configured as follows: If three consecutive warnings are issued and the RMSSD value does not improve, the wearable device will automatically initiate transcutaneous vagus nerve stimulation.
[0009] As a preferred technical solution, the intervention system further includes an associated early warning module, configured as follows: By performing time-series analysis on historical data and using a sliding window to uncover the temporal patterns of depressive episodes, a time-related warning is generated when the frequency of depression warnings in a certain time period exceeds 1.5 times the average level of the user.
[0010] As a preferred technical solution, the intervention system further includes an environmental correlation module, which generates time-related early warnings while analyzing environmental data, such as PM2.5, temperature, and humidity, and their correlation with HRV. When PM2.5 > 50 μg / m³ 3 If the SDNN value of HRV decreases by more than 15%, then an environment-related suggestion will be generated.
[0011] The present invention has the following beneficial effects: This invention is the first to precisely divide hard and soft thresholds according to frequency bands based on the spectral characteristics of PPG noise. By combining Sym8 wavelet basis + 5-level decomposition + hierarchical adaptive threshold, it achieves in-depth optimization of the depression monitoring scenario. In view of the high sensitivity of HRV nonlinear characteristics (ApEn / SampEn) to signal quality in depression monitoring, a hierarchical threshold strategy is customized and developed.
[0012] This invention effectively solves the problems of existing technologies that rely solely on HRV time-domain / frequency-domain parameters, neglecting nonlinear features such as approximate entropy and Poincaré diagrams, resulting in anxiety recognition accuracy of less than 80%, high latency of traditional algorithms, and lack of personalized adjustment mechanisms in biofeedback devices.
[0013] This invention innovatively employs HRV stress trajectory classification technology based on a latent category hybrid model to achieve individualized depression early warning.
[0014] This invention establishes an intelligent management mechanism for internet hospitals for the first time. It collects multi-dimensional physiological data of patients with depression in real time through wearable monitoring units. Based on the dynamic changes in the physiological data characteristics, it can be used as an auxiliary diagnostic tool for psychiatry, especially suitable for early screening and remote follow-up of depression, thereby reducing the consumption of medical resources. Attached Figure Description
[0015] Figure 1 This is a structural diagram of a depression monitoring system based on internet hospitals and heart rate variability, provided in an embodiment of the present invention.
[0016] Figure 2 The image shows the experimental results. Detailed Implementation
[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The embodiments of the depression state monitoring and intervention system based on Internet hospitals and heart rate variability in this invention include: A multimodal physiological signal acquisition module is used to simultaneously acquire raw heart rate variability signals, skin conductance fluctuations, and respiratory rate data. The signal processing module is used to preprocess the original signal and extract features to obtain multi-dimensional feature vectors; The depression level inference module constructs a multilayer perceptron neural network. The input is a multidimensional feature vector containing multidimensional features of heart rate variability, skin conductance fluctuation rate, and respiratory rate statistics. The output is the probability value of no depression, mild depression, and moderate to severe depression level.
[0019] An intervention system based on a depression state monitoring system, wherein the intervention system is configured as follows: By using a latent category mixture model to analyze the dynamic trajectory of users' HRV over the past 7 days, including data from resting state and daily activity state, users are divided into typical declining users whose LF / HF ratio decreases by ≥30% during stress and atypical rising users whose LF / HF ratio increases by ≥20% during stress. Based on user segmentation, intervention is carried out on users.
[0020] Specifically, the solution of the present invention is as follows: The hardware device of the system of this invention adopts a wrist-worn wearable terminal, with its shell made of medical-grade silicone material, measuring 45mm × 38mm × 12mm and weighing no more than 30g. The front of the device integrates a 0.96-inch OLED display screen for displaying real-time heart rate, depression level warnings, and other information; the back is equipped with a sensor array, including: The photoplethysmography (PPG) sensor uses dual-wavelength LEDs of 660nm and 940nm, with a sampling frequency set to 250Hz. The skin conductance response EDA electrode is made of Ag / AgCl material and has a sampling frequency of 10Hz. Triaxial accelerometer, measuring range ±8g, sampling frequency 50Hz; The device has a built-in percutaneous vagus nerve stimulation module. The stimulation electrode is located on the inside of the device close to the ulnar nerve at the wrist. It can output a square wave electrical signal with a frequency of 20Hz and a pulse width of 200μs. The intensity adjustment range is 0.5~2mA and the step size is 0.1mA.
[0021] The device's main control chip uses a low-power ARM Cortex-M4 processor, with built-in 128KB RAM and 1MB Flash, supports Bluetooth 5.0 and LTE-M dual-mode communication, has a battery capacity of 300mAh, and provides a battery life of ≥7 days in continuous monitoring mode. During the data acquisition phase, the device automatically enters real-time monitoring mode after power-on. The PPG sensor emits light signals to the skin of the wrist and receives the reflected light signals to obtain the raw photoplethysmography (PPG) wave signal. After analog-to-digital conversion, a digital signal is generated. Every 5 seconds, the peak detection algorithm identifies the R wave and outputs a set of RR interval sequences with a sequence length of 20 data points. Simultaneously, the EDA electrode collects the skin surface conductivity value in micro-Siemens, recording data every 100ms. The triaxial accelerometer collects the user's body movement information to eliminate abnormal physiological signals caused by motion interference. The respiratory rate is extracted from the baseline drift component of the PPG signal, and an adaptive filtering algorithm is used to separate the respiratory frequency component, updating the value every 1 second in breaths / minute.
[0022] In the signal preprocessing stage, the original PPG signal was decomposed into 5 levels using the Sym8 wavelet basis. After decomposition, hard thresholding was applied to the high-frequency coefficients of levels 1-3 to remove electromyographic interference, while soft thresholding was applied to the low-frequency coefficients of levels 4-5 to eliminate baseline drift. The threshold calculation adopted a heuristic threshold formula. The processed signal was then reconstructed into a denoised pulse wave signal using wavelet reconstruction. For the RR interval sequence, linear interpolation was used to resample it to a 1Hz isochronous sequence, and then a 5-point moving average filter was applied to remove transient noise, finally obtaining stable HRV time series data. During feature extraction, based on the preprocessed HRV data, multi-dimensional features are calculated through the following steps: In the time domain, the standard deviation of the normal RR interval (SDNN) is obtained by calculating the standard deviation of all RR intervals within a 5-minute window; the root mean square (RMSSD) of the difference between adjacent RR intervals is obtained by taking the square root of the average of the squares of the differences between adjacent RR intervals. In the frequency domain, fast Fourier transform is used to estimate the power spectrum of the resampled HRV sequence. The low-frequency power (LF) is the power integral of the 0.04–0.15 Hz band, and the high-frequency power (HF) is the power integral of the 0.15–0.4 Hz band. The LF / HF ratio is the ratio of their power values. In the nonlinear features, the approximate entropy (ApEn) is calculated using an embedding dimension of 2 and a similarity tolerance of 0.2 times the standard deviation. The sample entropy (SampEn) uses the same embedding dimension and tolerance parameters. The short axis (SD1) and long axis (SD2) of the Poincaré plot are obtained by calculating the scatter distribution characteristics of the RR interval sequence. SD1 reflects short-term heart rate variability, and SD2 reflects long-term heart rate variability. Meanwhile, the skin conductance response variability was obtained by calculating the ratio of the standard deviation to the mean of the EDA data within a 1-minute window, and the respiratory rate characteristic was the ratio of the mean to the standard deviation within a 5-minute window.
[0023] During the model inference phase, the extracted HRV multidimensional features, skin conductance fluctuation rate, and respiratory rate features are combined into a multidimensional feature vector, which is then input into a pre-trained multilayer perceptron (MLP) neural network. This network consists of an input layer, three hidden layers (64, 32, and 16 neurons respectively, using ReLU activation), and an output layer (three neurons corresponding to no depression, mild depression, and moderate to severe depression, using Softmax activation). The model is loaded into the edge computing chip of the wearable device, employing fixed-point arithmetic optimization to reduce power consumption. Feature vector input and inference calculations are performed every 10 seconds, outputting the probability value for each depression level. When the probability of severe depression is >0.7, the device displays a red warning icon on the OLED screen and triggers a vibration module for alert; when the probability of mild depression is >0.6, a yellow warning icon is displayed, and no vibration is triggered.
[0024] During personalized intervention, the system first classifies users' HRV stress response patterns based on the Latent Class Mixture Model (LCMM). This model analyzes users' HRV dynamic trajectories over a 7-day history, including data from resting and daily activities, categorizing users into typical declining types (LF / HF ratio decreases by ≥30% during stress) and atypical rising types (LF / HF ratio increases by ≥20% during stress). For typical declining users, when a depression warning is triggered, the mobile app automatically pushes a deep breathing guided course. The course uses a rhythm of 4 seconds of inhalation, 4 seconds of breath-holding, and 6 seconds of exhalation, guided synchronously through screen animations and voice prompts. Each training session lasts 5 minutes, during which the RMSSD value of HRV is monitored in real time. When this value rises to more than 90% of the baseline level (average of the resting state over the past 3 days), the training automatically ends. For users with atypical elevated muscle tone, progressive muscle relaxation training is initiated after an alert is triggered. The app guides users to alternately tense and relax muscles in the following order: hand-forearm-upper arm-shoulder-face-trunk-lower limbs. Each muscle is tensed for 5 seconds and relaxed for 10 seconds, for a total of 10 minutes. Skin conductance is monitored during the training. Training is terminated when the conductance drops below 80% of the pre-alert level. If three consecutive alerts are triggered and the HRV index does not improve, the wearable device automatically initiates transcutaneous vagus nerve stimulation. The initial intensity is set to 1 mA, and it is adjusted every 30 seconds based on the real-time HF power value. The current intensity is maintained when the HF power increases by 15% compared to before stimulation; otherwise, it is increased by 0.1 mA. Stimulation is paused after 5 minutes. If the user is still in an alert state after 3 minutes, stimulation is repeated. In terms of data management, wearable devices synchronize hourly HRV characteristic data, depression level results, and intervention records to a mobile app via Bluetooth 5.0. The app then uploads the data to a cloud database via an LTE-M network. The cloud uses a MySQL database to store the data. Each record includes a user ID, timestamp, 12 HRV characteristics, 2 auxiliary physiological characteristics, depression level, intervention type, and post-intervention indicator changes. The cloud server uses an LSTM network to perform time-series analysis on historical data. A sliding window with a 7-day window size is used to mine the temporal patterns of depressive episodes. When the frequency of depression warnings in a certain time period exceeds 1.5 times the user's average level, a time-related warning is generated. Simultaneously, the correlation between environmental data (PM2.5, temperature, humidity) and HRV is analyzed. When PM2.5 > 50 μg / m³... 3 If the HRV's SDNN value drops by more than 15%, an environment-related suggestion will be generated and pushed to the user via the app.
[0025] The clinical validation of this invention was conducted in the Department of Psychiatry and Psychology, South China Hospital Affiliated to Shenzhen University. A total of 120 subjects aged 18-65 years were selected and divided into four groups according to their mental status: a healthy control group (without anxiety or depression) of 30 cases, a simple anxiety group of 30 cases, a simple depression group of 30 cases, and a comorbid anxiety and depression group of 30 cases; all subjects in each group with anxiety / depression met the corresponding diagnostic criteria of the International Classification of Diseases (ICD-11).
[0026] The experiment lasted for 8 weeks. The system of this invention was used twice daily (morning and evening, 30 minutes each time) to collect multidimensional heart rate variability (HRV) characteristics, skin conductance, and respiratory rate data from all subjects in real time. The Self-Rating Depression Scale (SDS), Hamilton Depression Rating Scale (HAMD-17), Self-Rating Anxiety Scale (SAS), and Hamilton Anxiety Rating Scale (HAMA) were used as control standards. A psychiatrist assessed the scales weekly, and the overall evaluation results of the scales were compared with the results identified by the system of this invention for consistency analysis.
[0027] Verification showed that the system of this invention achieved an accuracy rate of 89.3% in identifying individuals without anxiety or depression, and 74.9% in identifying individuals with moderate to severe anxiety and depression comorbidities. Compared to traditional devices using a single HRV feature (parallel controlled experiment at the same institution, n=30), this invention improves the combined accuracy rate of identifying anxiety and depression by 12%–18%. Furthermore, in a 6-week follow-up study of 15 subjects in the moderate to severe anxiety and depression comorbidity group who received the personalized intervention program of this invention (relaxation training and lifestyle guidance dynamically adjusted based on real-time physiological data), the time required for their HRV indicators to recover to healthy baseline levels was approximately 40% shorter than that of the control subjects who received conventional intervention.
[0028] The experimental comparison data are shown in Table 1 and Figure 2 : Table 1
[0029] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A depression monitoring system based on internet hospitals and heart rate variability, characterized in that, The system includes: A multimodal physiological signal acquisition module is used to simultaneously acquire raw heart rate variability signals, skin conductance fluctuations, and respiratory rate data. The signal processing module is used to preprocess the original signal and extract features to obtain multi-dimensional feature vectors; The depression level inference module constructs a multilayer perceptron neural network, takes a multidimensional feature vector containing multidimensional features of heart rate variability, skin conductance fluctuation rate and respiratory rate statistics as input, and outputs probability values of no depression, mild depression and moderate to severe depression level. Preprocessing of the raw heart rate variability signal includes: The raw heart rate variability signal was decomposed into a 5-level wavelet based on the Sym8 wavelet basis. After decomposition, the high-frequency coefficients of the first to third levels were processed by hard thresholding to remove electromyographic interference, and the low-frequency coefficients of the fourth to fifth levels were processed by soft thresholding to eliminate baseline drift. The processed signal was then reconstructed by wavelet to restore the denoised pulse wave signal. For the RR interval sequence, linear interpolation is used to resample it to a 1Hz isochronous sequence, and then a 5-point moving average filter is used to remove instantaneous noise, finally obtaining stable HRV time series data; The feature extraction includes: Based on the preprocessed HRV time series data, multi-dimensional features are calculated through the following steps: The standard deviation of normal RR intervals (SDNN) is obtained by calculating the standard deviation of all RR intervals within a 5-minute window. The root mean square (RMSSD) of the difference between adjacent RR intervals is obtained by taking the square root of the average of the squared differences of the differences between adjacent RR intervals. The power spectrum of the resampled HRV sequence is estimated by using Fast Fourier Transform (FFT) to calculate the low-frequency power (LF), high-frequency power (HF), and the LF / HF ratio. The low-frequency power (LF) is the power integral of the 0.04–0.15 Hz frequency band, the high-frequency power (HF) is the power integral of the 0.15–0.4 Hz frequency band, and the LF / HF ratio is the ratio of the power values of the two. The approximate entropy ApEn was calculated using a parameter setting of embedding dimension 2 and similarity tolerance of 0.2 times the standard deviation. The sample entropy SampEn was obtained using the same embedding dimension and tolerance parameter. The short axis SD1 and long axis SD2 of the Poincaré plot were obtained by calculating the scatter distribution characteristics of the RR interval sequence. The skin conductance response volatility is obtained by calculating the ratio of the standard deviation to the mean of the EDA data within a 1-minute window. The mean and standard deviation of respiratory rate within a 5-minute window are used as respiratory rate characteristics.
2. The system according to claim 1, characterized in that, In the depression level inference module, the multilayer perceptron neural network includes an input layer, three hidden layers, and an output layer; the three hidden layers have 64, 32, and 16 neurons respectively, and the activation function is ReLU; the output layer includes three neurons, corresponding to no depression, mild depression, and moderate to severe depression, and the activation function is Softmax.
3. An intervention system based on the depression monitoring system of claim 1, characterized in that, The intervention system is configured as follows: By using a latent category mixture model to analyze the dynamic trajectory of users' HRV over the past 7 days, including data from resting state and daily activity state, users are divided into typical declining users whose LF / HF ratio decreases by ≥30% during stress and atypical rising users whose LF / HF ratio increases by ≥20% during stress. Based on user segmentation, intervention is carried out on users.
4. The intervention system according to claim 3, characterized in that, The intervention system also includes a wearable device and a personalized intervention module, wherein the personalized intervention module is configured as follows: For typical declining users, when a depression warning is triggered, a course is pushed to them. The course uses an inhalation-breath-exhalation rhythm and is guided synchronously through screen animations and voice prompts. During the course, the RMSSD value of HRV is monitored in real time. The training ends when the RMSSD value rises back to more than 90% of the baseline level.
5. The intervention system according to claim 4, characterized in that, The personalized intervention module is configured as follows: For users with atypical elevated muscle type, progressive muscle relaxation training is initiated after the warning is triggered. Users are guided to alternately tense and relax their muscles in the following order: hand-forearm-upper arm-shoulder-face-trunk-lower limb. Skin conductance values are monitored during the training. The training ends when the skin conductance value drops to below 80% of the pre-warning level.
6. The intervention system according to claim 5, characterized in that, The personalized intervention module is configured as follows: If three consecutive warnings are issued and the RMSSD value does not improve, the wearable device will initiate transcutaneous vagus nerve stimulation.
7. The intervention system according to claim 6, characterized in that, The intervention system also includes a correlation early warning module, configured as follows: By performing time-series analysis on historical data and using a sliding window to uncover the temporal patterns of depressive episodes, a time-related warning is generated when the frequency of depression warnings in a certain time period exceeds 1.5 times the average level of the user.
8. The intervention system according to claim 7, characterized in that, The intervention system also includes an environmental correlation module, which generates time-related early warnings while analyzing environmental data, such as PM2.5, temperature, and humidity, and their correlation with HRV. When PM2.5 > 50 μg / m³ 3 If the SDNN value of HRV decreases by more than 15%, then an environment-related suggestion will be generated.