Fall early warning method and system for sleep quality analysis based on big data

Abstract

The invention provides a fall early warning method and system for sleep quality analysis based on big data, and relates to the field of health data processing, the method comprises the following steps: obtaining a big data sample, and determining a plurality of user types, sleep quality factors associated with each user type and weights of the sleep quality factors; for each user type, establishing a fall risk prediction model corresponding to the user type; determining the user type of the current user; acquiring sleep monitoring data of a current user through sleep monitoring equipment; based on the sleep quality factor associated with the user type of the current user, extracting sleep quality features of the current user from the sleep monitoring data of the current user; through the fall risk prediction model corresponding to the user type of the current user, the fall risk of the current user is predicted based on the sleep quality characteristics of the current user, and the method has the advantage of realizing early warning of the fall risk.

Description

Technical Field

[0001] The present invention relates to the field of health data processing, and in particular to a fall warning method and system for sleep quality analysis based on big data. Background Art

[0002] Elderly individuals who sleep too little (e.g., ≤4 hours / night) have a significantly increased incidence of falls and fall-related fractures. Insufficient sleep can lead to decreased physical function in older adults, including decreased muscle strength, balance, and reaction time, which can increase the risk of falls. Furthermore, sleep disorders can cause older adults to wake up frequently during the night, affecting sleep quality and increasing the risk of falls.

[0003] Existing fall warning technologies only focus on detecting fall events, but ignore the risk assessment before a fall. Big data-based methods can comprehensively analyze multi-dimensional data such as sleep quality, physical activity, and physiological indicators to comprehensively assess fall risks and take preventive measures in advance. Therefore, it is necessary to provide a fall warning method and system based on big data sleep quality analysis to achieve early warning of fall risks. Summary of the Invention

[0004] The present invention provides a fall warning method for sleep quality analysis based on big data, comprising: obtaining a big data sample, determining multiple user types, a sleep quality factor associated with each user type, and a weight of the sleep quality factor; for each user type, establishing a fall risk prediction model corresponding to the user type; determining the user type of the current user; obtaining sleep monitoring data of the current user through a sleep monitoring device; extracting the sleep quality characteristics of the current user from the sleep monitoring data of the current user based on the sleep quality factor associated with the user type of the current user; and predicting the fall risk of the current user based on the sleep quality characteristics of the current user through the fall risk prediction model corresponding to the user type of the current user.

[0005] Furthermore, the big data sample includes medical records, sleep monitoring data and fall risk assessment data of multiple sample users: determining multiple user types, including: determining multiple diseases to be screened and multiple drugs to be screened; determining the influence coefficient of each disease to be screened on the fall risk based on the medical records and fall data of multiple sample users, and screening influencing diseases from multiple diseases to be screened based on the influence coefficient of each disease to be screened on the fall risk; determining the influence coefficient of each drug to be screened on the fall risk based on the medical records and fall data of multiple sample users, and screening influencing drugs from multiple drugs to be screened based on the influence coefficient of each drug to be screened on the fall risk; determining multiple user types based on the influencing diseases, influencing drugs and the medical records of multiple sample users.

[0006] Furthermore, based on the influencing diseases, influencing drugs and medical records of multiple sample users, multiple user types are determined, including: for each sample user, determining the influencing disease identification vector of the sample user based on the sample user's medical record information and influencing diseases, and determining the influencing drug identification vector of the sample user based on the sample user's medical record information and influencing drugs; determining the similarity of the identification vectors of any two sample users based on the influencing disease identification vectors and influencing drug identification vectors; and dividing the multiple sample users into multiple user clusters based on the similarity of the identification vectors of any two sample users through a clustering algorithm, each user cluster corresponding to a user type.

[0007] Furthermore, determining the sleep quality factor associated with each user type includes: determining multiple sleep quality factors to be screened and multiple fall risk assessment indicators; for each user type, based on the sleep monitoring data and fall risk assessment data of multiple sample users included in the user cluster, determining the correlation coefficient between each sleep quality factor to be screened and each fall risk assessment indicator, and determining the sleep quality factor associated with the user type based on the correlation coefficient between each sleep quality factor to be screened and each fall risk assessment indicator.

[0008] Furthermore, the multiple sleep quality factors to be screened include at least the number of awakenings, total sleep time, first stage sleep time and second stage sleep time; the multiple fall risk assessment indicators include at least vestibular nerve signal conduction efficiency, creatine kinase level, melatonin secretion cycle and systolic blood pressure.

[0009] Furthermore, determining the weight of the sleep quality factor associated with the user type includes: determining the weight of the sleep quality factor associated with the user type according to a correlation coefficient between each sleep quality factor associated with the user type and each fall risk assessment indicator.

[0010] Furthermore, the sleep monitoring device includes a millimeter wave radar component and a heart rate monitoring component.

[0011] Furthermore, based on the sleep quality factor associated with the user type of the current user, the sleep quality characteristics of the current user are extracted from the sleep monitoring data of the current user, including: for each user type, based on the sleep quality factor associated with the user type, establishing a feature extraction model corresponding to the user type; and extracting the sleep quality characteristics of the current user from the sleep monitoring data of the current user through the feature extraction model corresponding to the user type of the current user.

[0012] Furthermore, the method also includes: if the current user's fall risk is greater than a preset fall risk threshold, determining an environment control strategy and exercise recommendations based on the current user's sleeping environment information and the current user's sleep quality characteristics.

[0013] The present invention provides a fall warning system for sleep quality analysis based on big data, and applies the above-mentioned fall warning method for sleep quality analysis based on big data, including: a big data analysis module, used to obtain big data samples, determine multiple user types, sleep quality factors associated with each user type, and the weights of the sleep quality factors; a model establishment module, used to establish a fall risk prediction model corresponding to the user type for each user type; a type determination module, used to determine the user type of the current user; a sleep monitoring module, used to obtain sleep monitoring data of the current user; a feature extraction module, used to extract the sleep quality features of the current user from the sleep monitoring data of the current user based on the sleep quality factors associated with the user type of the current user; and a risk prediction module, used to predict the fall risk of the current user based on the sleep quality features of the current user through the fall risk prediction model corresponding to the user type of the current user.

[0014] Compared with the existing technology, the fall warning method and system based on big data sleep quality analysis provided by the present invention have at least the following beneficial effects: By distinguishing different user types and establishing a corresponding fall risk prediction model for each user type, personalized fall warning is achieved, improving the accuracy and pertinence of the warning.

[0015] Incorporating sleep quality factors into fall risk prediction takes into account the potential impact of sleep on fall risk, making the early warning system more comprehensive and scientific.

[0016] Based on big data samples, user types, sleep quality factors and their weights are determined, which provides a solid data foundation for early warning and improves its reliability and effectiveness.

[0017] The user's sleep monitoring data can be obtained in real time through sleep monitoring equipment, and sleep quality characteristics can be extracted based on this data to predict the risk of falls in real time, which helps to detect and prevent falls in a timely manner.

[0018] Effective fall warnings can help users (especially the elderly or those in poor health) take preventive measures, reduce the harm caused by falls, and thus improve their quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] This specification will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, like numbers represent like structures, wherein: Figure 1 is a flowchart of a fall warning method based on sleep quality analysis based on big data according to some embodiments of this specification; Figure 2This is a module diagram of a fall warning system for sleep quality analysis based on big data according to some embodiments of this specification. DETAILED DESCRIPTION

[0020] To more clearly illustrate the technical solutions of the embodiments of this specification, the following briefly describes the drawings required for describing the embodiments. Obviously, the drawings described below are merely examples or embodiments of this specification. Those skilled in the art can apply this specification to other similar scenarios based on these drawings without inventive effort. Unless otherwise apparent from the context or otherwise noted, the same reference numerals in the figures represent the same structure or operation.

[0021] First, experiments were combined to demonstrate the association between sleep quality and fall risk.

[0022] Sleep monitoring and fall data were collected from 300 elderly people aged 65-90 years across three nursing homes between January 2024 and March 2025. Hospital A housed single rooms in a quiet environment, Hospital B housed double rooms in a standard environment, and Hospital C housed a natural sound environment, with 100 residents each. All residents had no severe cognitive impairment, could walk independently, and volunteered to participate in the monitoring.

[0023] Sleep monitoring: Use millimeter-wave radar to continuously record total sleep time, deep sleep percentage, and number of awakenings at night.

[0024] Fall records: Nurses record fall events (including time, location, and degree of injury) daily to exclude involuntary falls.

[0025] Logistic regression was used to analyze the association between sleep quality indicators and fall risk, and odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. The chi-square test was used to compare fall rates among different nursing homes.

[0026] The results of the logistic regression analysis of sleep quality indicators and fall risk are shown in Table 1.

[0027] Table 1 Sleep quality indicators Fall incidence OR value (95% CI) P-value Total sleep time < 6 hours 32.6% 3.12（1.89-5.14） <0.001 Deep sleep ratio <15% 28.4% 2.76（1.67-4.55） 0.002 Nocturnal awakenings ≥ 3 times / night 37.1% 4.05（2.33-7.02） <0.001 Table 1 shows that the incidence of falls was significantly increased in elderly individuals with total sleep duration <6 hours, deep sleep <15%, and ≥3 nighttime awakenings per night, with ORs of 3.12 (95% CI: 1.89-5.14), 2.76 (95% CI: 1.67-4.55), and 4.05 (95% CI: 2.33-7.02), respectively, all with P values ​​<0.05. The number of nighttime awakenings had the greatest impact on the risk of falls (OR = 4.05).

[0028] The fall rates in different nursing homes are shown in Table 2 .

[0029] Table 2 mechanism Average deep sleep time Number of falls / 100 person-years Proportion of severe injuries Hospital A 1.8±0.4 hours 18.2 12.1% Hospital B 1.2±0.6 hours 42.7 27.6% C Hospital 2.1±0.3 hours 9.5 5.3% Table 2 shows that Hospital C (natural sound environment) had the longest average deep sleep duration (2.1 ± 0.3 hours), the lowest number of falls per 100 person-years (9.5), and the lowest proportion of severe injuries (5.3%). A significance test showed that the fall rate in Hospital C was significantly lower than that in Hospital A (18.2 falls per 100 person-years) and Hospital B (42.7 falls per 100 person-years) (χ² = 15.32, P < 0.001).

[0030] Impact mechanism analysis: 1. Physiological pathway: Insufficient deep sleep leads to decreased vestibular function (balance ability decreases by 23.7%) and increased daytime fatigue by 41.2%.

[0031] Nighttime awakenings induce orthostatic hypotension (a drop in systolic blood pressure >20 mmHg increases the risk by 2.3 times).

[0032] 2. Environmental Interaction Noise >50 decibels increased the number of awakenings by 1.8 times (Hospital B vs Hospital C, P=0.008).

[0033] Single rooms reduce obstacles to toileting at night (the fall rate in Hospital A was 57.4% lower than that in Hospital B).

[0034] Experiments have confirmed that sleep quality is a significant factor influencing fall risk in the elderly. The number of nighttime awakenings has the greatest impact on fall risk (OR = 4.05), possibly due to the decreased balance and delayed reflexes caused by nighttime awakenings. Insufficient deep sleep indirectly increases fall risk by affecting vestibular function and daytime fatigue, consistent with previous research.

[0035] Figure 1 This is a flow chart of a fall warning method based on sleep quality analysis based on big data according to some embodiments of this specification, such as Figure 1 As shown, the fall warning method based on big data sleep quality analysis can include the following steps.

[0036] Step 110 , obtaining a large data sample, determining multiple user types, a sleep quality factor associated with each user type, and a weight of the sleep quality factor.

[0037] Among them, the big data samples include medical records information, sleep monitoring data and fall risk assessment data of multiple sample users. Specifically, medical record information may include the user's age, gender, medical history (such as chronic diseases, surgical history, etc.), medication status, etc. Sleep monitoring data is obtained through technical means such as wearable devices, mattress sensors or millimeter wave radars. The sleep monitoring data continuously records the user's total sleep time, deep sleep percentage, number of awakenings at night and other sleep quality indicators. The fall risk assessment data may include multiple fall risk assessment values, among which the multiple fall risk assessment indicators include at least vestibular nerve signal conduction efficiency, creatine kinase level, melatonin secretion cycle and systolic blood pressure.

[0038] In some embodiments, determining multiple user types includes: Identify multiple diseases and drugs to be screened. Diseases to be screened are those closely related to the risk of falls, such as Parkinson's disease, stroke sequelae, osteoporosis, cardiovascular disease, etc. Drugs to be screened are drugs that may affect sleep quality or increase the risk of falls, such as sedatives, antidepressants, and antihypertensive drugs; Based on the medical records and fall data of multiple sample users, determine the impact coefficient of each disease to be screened on the risk of falls. Based on the impact coefficient of each disease to be screened on the risk of falls, screen the influencing diseases from the multiple diseases to be screened. Specifically, use statistical methods (such as logistic regression, Cox proportional hazards model, etc.) to analyze the impact coefficient of each disease to be screened on the risk of falls. Based on the size of the impact coefficient and the significance level (such as P value <0.05), screen the influencing diseases that have a significant impact on the risk of falls from the multiple diseases to be screened. Based on the medical records and fall data of multiple sample users, determine the impact coefficient of each drug to be screened on the risk of falls. Based on the impact coefficient of each drug to be screened on the risk of falls, screen the influencing drugs from the multiple drugs to be screened. Specifically, use statistical methods (such as logistic regression, Cox proportional hazards model, etc.) to analyze the impact coefficient of each drug to be screened on the risk of falls. Based on the size of the impact coefficient and the significance level (such as P value <0.05), screen the influencing diseases that have a significant impact on the risk of falls from the multiple drugs to be screened. Based on influencing diseases, influencing drugs and medical record information of multiple sample users, multiple user types are determined.

[0039] In some embodiments, based on the influencing diseases, influencing drugs, and medical history information of multiple sample users, multiple user types are determined, including: For each sample user, determine the sample user's influencing disease identification vector based on the sample user's medical record information and influencing diseases; determine the sample user's influencing drug identification vector based on the sample user's medical record information and influencing drugs; Determine the similarity of the identification vectors of the two sample users based on the identification vectors of the affected diseases and the identification vectors of the affected drugs of any two sample users; Through the clustering algorithm, multiple sample users are divided into multiple user clusters according to the similarity of the identification vectors of any two sample users, and each user cluster corresponds to a user type.

[0040] Specifically, for each sample user, their medical history is checked to see if they have these influencing diseases. A binary vector (identification vector) is constructed with a length equal to the total number of influencing diseases. If a user has a certain influencing disease, the corresponding position is marked as 1; otherwise, it is marked as 0. For example, if there are three influencing diseases (A, B, and C), and a user has diseases A and C, their influencing disease identification vector is [1, 0, 1].

[0041] For each sample user, check their medical records to see if they are currently using these influencing medications. A binary vector (identification vector) is constructed with a length equal to the total number of influencing medications. If a user is currently using a particular influencing medication, the corresponding position is marked as 1; otherwise, it is marked as 0. For example, if there are two influencing medications (X, Y), and a user is currently using medication X, then their influencing medication identification vector is [1, 0].

[0042] For any two sample users, the cosine similarity of their disease-affecting identification vectors and the cosine similarity of their drug-affecting identification vectors are calculated respectively. The cosine similarity of the disease-affecting identification vectors and the drug-affecting identification vectors are weighted summed to obtain the identification vector similarity of the two sample users.

[0043] The K-means clustering algorithm can be used to divide multiple sample users into multiple user clusters based on the similarity of the identification vectors of any two sample users. The specific steps include: S11. Initialize cluster center: Randomly select K sample users as initial cluster centers.

[0044] S12. Assign samples to clusters: For each sample user, the distance (e.g., Euclidean distance) between its comprehensive identification vector and all cluster centers is calculated.

[0045] Assign samples to the cluster with the closest distance.

[0046] S13. Update cluster center: Recalculate the center of each cluster (i.e., the mean of all sample identity vectors within the cluster).

[0047] S14, iterative optimization: Repeat steps S12 and S13 until the cluster center no longer changes significantly or the maximum number of iterations is reached.

[0048] For each user cluster, the distribution of the influencing diseases and influencing drugs of the users in the cluster is counted.

[0049] Example: User cluster 1: 80% of users suffer from Parkinson's disease and 60% of users are taking antidepressants.

[0050] User cluster 2: 70% of users suffer from osteoporosis, with no significant effect from medication.

[0051] Name each user cluster based on the common characteristics of the users in the cluster.

[0052] Example: User cluster 1 is named “Parkinson's disease combined with depression patient type”.

[0053] User cluster 2 is named “osteoporosis patient type”.

[0054] In some embodiments, determining a sleep quality factor associated with each user type includes: Determining multiple sleep quality factors to be screened and multiple fall risk assessment indicators, wherein the multiple sleep quality factors to be screened include at least the number of awakenings, total sleep time, first stage sleep time, and second stage sleep time; and the multiple fall risk assessment indicators include at least vestibular nerve signal conduction efficiency, creatine kinase level, melatonin secretion cycle, and systolic blood pressure; For each user type, based on the sleep monitoring data and fall risk assessment data of multiple sample users included in the user cluster, the correlation coefficient between each sleep quality factor to be screened and each fall risk assessment indicator is determined. According to the correlation coefficient between each sleep quality factor to be screened and each fall risk assessment indicator, the sleep quality factor associated with the user type is determined.

[0055] Specifically, the number of awakenings is the number of awakenings during sleep at night, which reflects the continuity of sleep. Awakening is defined as body movement >3 times / hour or heart rate >100bpm for 5 minutes.

[0056] Total sleep time: The total time of sleep at night, reflecting the adequacy of sleep.

[0057] Duration of the first stage of sleep: The time spent in the light sleep stage usually accounts for less than 50% of the total sleep time. It is defined as the first stage of sleep when the brain waves are dominated by theta waves (4-7Hz) and the breathing rate is ≤12 times / minute.

[0058] Duration of stage 2 sleep: The time in the deep sleep stage is crucial for the body's recovery and repair, usually accounting for 15%-25% of the total sleep time. It is defined as stage 2 sleep when the brain wave delta wave (0.5-4Hz) is dominant and the body movement frequency is ≤1 time / hour.

[0059] Vestibular nerve signal conduction efficiency: reflects the function of the vestibular system, which is related to balance and spatial positioning. Decreased function may increase the risk of falls.

[0060] Creatine kinase level: reflects muscle damage or metabolic status. Decreased muscle function may affect walking stability.

[0061] Melatonin secretion cycle: reflects the biological clock and sleep rhythm. Disruption may lead to sleep disorders and increased risk of falls.

[0062] Systolic blood pressure: Fluctuations in blood pressure may affect blood supply to the brain, causing dizziness or balance problems and increasing the risk of falls.

[0063] From the user cluster corresponding to each user type, extract the sleep monitoring data and fall risk assessment data for all sample users. The data for each sample user should include: measured values ​​for sleep quality factors to be screened (e.g., number of awakenings, total sleep duration, etc.); and measured values ​​for fall risk assessment indicators (e.g., vestibular nerve signal conduction efficiency, creatine kinase levels, etc.). Substitute the measured values ​​of the sleep quality factors to be screened and the fall risk assessment indicators for the user cluster corresponding to the user type into the Spearman rank correlation coefficient to calculate the correlation coefficient between the sleep quality factors to be screened and the fall risk assessment indicators.

[0064] According to the significance and size of the correlation coefficient, the sleep quality factor that is significantly correlated with the fall risk assessment index is screened out as the sleep quality factor associated with the user type.

[0065] Significance: The P value of the correlation coefficient is less than the set significance level (such as 0.05).

[0066] Correlation strength: According to research needs, set the absolute value threshold of the correlation coefficient (such as |r|>0.3).

[0067] Screening Process: For each user type, the correlation coefficients of all sleep quality factors to be screened and fall risk assessment indicators are traversed.

[0068] The sleep quality factors corresponding to the correlation coefficients that meet the screening criteria are screened out.

[0069] Example: In the "Parkinson's disease combined with depression patient type", it was found that the "number of awakenings" was significantly negatively correlated with the "vestibular nerve signal conduction efficiency" (r = -0.45, P<0.01), so the "number of awakenings" is the sleep quality factor associated with this user type.

[0070] In the "osteoporosis patient type", it was found that "total sleep duration" was significantly positively correlated with "creatine kinase level" (r = 0.38, P<0.05), so "total sleep duration" is the sleep quality factor associated with this user type.

[0071] In some embodiments, determining the weight of the sleep quality factor associated with the user type includes: The weight of the sleep quality factor associated with the user type is determined according to the correlation coefficient between each sleep quality factor associated with the user type and each fall risk assessment indicator.

[0072] Specifically, for each sleep quality factor associated with the user type, the correlation coefficients between the sleep quality factor and each fall risk assessment indicator are averaged to obtain the correlation coefficient mean.

[0073] The correlation coefficient means of each sleep quality factor associated with the user type are summed to obtain the sum of the correlation coefficient means, and the ratio of the correlation coefficient of the sleep quality factor associated with the user type to the sum of the correlation coefficient means is used as the weight of the sleep quality factor associated with the user type.

[0074] Step 120: For each user type, establish a fall risk prediction model corresponding to the user type.

[0075] Specifically, the fall risk prediction model corresponding to the user type can be a deep learning model.

[0076] There are significant differences in the fall risk mechanisms of different user types, and it is necessary to build a personalized fall risk prediction model based on their unique sleep quality factors and risk characteristics.

[0077] Deep learning models (such as long short-term memory networks, Transformer, convolutional neural networks, etc.) are suitable for processing high-dimensional, nonlinear, and time series data, and can capture the complex relationship between sleep quality factors and fall risk.

[0078] Input and output design of the fall risk prediction model: 1. Input layer: Time series data: sleep quality factors for multiple consecutive days (e.g., 7 days × 5 factors = 35-dimensional features) and the weights of sleep quality factors associated with user types.

[0079] 2. Output layer: Binary classification: fall risk (0 = low risk, 1 = high risk).

[0080] Regression: Fall risk probability (0-1) or prediction of future falls.

[0081] 3. Model training and optimization Data preprocessing: Normalization: Z-score normalization is performed on the sleep quality factor.

[0082] Sliding window: Split multiple consecutive days of data into sequences of fixed length (such as 7 days as a window).

[0083] Data augmentation: Adding Gaussian noise or random occlusion to time series data to improve generalization capabilities.

[0084] 4. Model training: Loss function: binary cross entropy (classification) or mean squared error (regression).

[0085] Optimizer: Adam (adaptive learning rate).

[0086] Regularization: Dropout (to prevent overfitting), L2 regularization.

[0087] 5. Hyperparameter tuning: Tune the following parameters using grid search or Bayesian optimization: Number of LSTM layers (1–3).

[0088] Number of hidden units (32~256).

[0089] Learning rate (1e-3~1e-5).

[0090] Evaluation Metrics: Classification tasks: accuracy, recall, F1 score, ROC-AUC.

[0091] Regression tasks: mean square error (MSE), mean absolute error (MAE).

[0092] Step 130: Determine the user type of the current user.

[0093] Specifically, the user type of the current user may be determined based on the user's medical record information.

[0094] Step 140: Obtain the sleep monitoring data of the current user through the sleep monitoring device.

[0095] Among them, sleep monitoring equipment includes millimeter wave radar components and heart rate monitoring components.

[0096] Millimeter-wave radar components Working principle: Millimeter-wave radar analyzes the distance, speed, and angle of the target object by emitting high-frequency electromagnetic waves (usually 24GHz or 77GHz) and receiving reflected signals.

[0097] In sleep monitoring, radar can penetrate light fabrics (such as sheets) and obtain physiological and movement information without direct contact with the user.

[0098] Function and data collection: Respiratory monitoring: Detects respiratory rate and depth through chest micro-movements.

[0099] Body movement monitoring: Captures large movements such as turning over and sitting up to assess the degree of sleep fragmentation.

[0100] Micro-movement analysis: Identifies subtle limb tremors or twitches to assist in determining rapid eye movement (REM) sleep or abnormal sleep behaviors (such as RBD, rapid eye movement sleep behavior disorder).

[0101] Spatial positioning: monitors changes in the user's body position during sleep (such as supine or side sleeping).

[0102] Heart rate monitoring component Working principle: Measure the user's heart rate and heart rate variability (HRV) using photoplethysmography (PPG) or electrocardiogram (ECG) technology.

[0103] PPG sensors are usually integrated into wearable devices (such as bracelets, chest straps) or mattress sensors, and detect blood volume changes through LED light and photodiodes.

[0104] Function and data collection: Heart rate monitoring: Real-time recording of resting heart rate and heart rate changes during sleep.

[0105] Step 150 : extracting a sleep quality feature of the current user from the sleep monitoring data of the current user based on the sleep quality factor associated with the user type of the current user.

[0106] In some embodiments, step 150 specifically includes: For each user type, a feature extraction model corresponding to the user type is established based on the sleep quality factor associated with the user type; The sleep quality features of the current user are extracted from the sleep monitoring data of the current user through a feature extraction model corresponding to the user type of the current user.

[0107] Specifically, in fall risk prediction, the mechanisms by which sleep quality factors influence fall risk vary significantly across user types. Therefore, a customized feature extraction model is needed for each user type to extract features highly correlated with fall risk from the raw sleep monitoring data. This feature extraction model can be a deep learning model.

[0108] Input data preprocessing: Standardize, normalize, or perform sliding window segmentation on the current user's sleep monitoring data.

[0109] Example: Divide the millimeter-wave radar micro-motion signal into 1-minute windows and calculate the micro-motion energy in each window.

[0110] Feature extraction: The preprocessed data is input into the feature extraction model corresponding to the user type and the sleep quality features are output.

[0111] Step 160 : Predicting the fall risk of the current user based on the sleep quality characteristics of the current user by using a fall risk prediction model corresponding to the user type of the current user.

[0112] Specifically, the sleep quality characteristics of the current user may be input into a fall risk prediction model corresponding to the user type of the current user to predict the fall risk of the current user.

[0113] In some embodiments, the method further includes step 170 , if the current user's fall risk is greater than a preset fall risk threshold, determining an environment control strategy and exercise recommendations based on the current user's sleeping environment information and the current user's sleep quality characteristics.

[0114] Specifically, through correlation analysis or causal reasoning models, the association between environmental factors and sleep quality characteristics is established to identify key risk factors.

[0115] Based on the results of the association analysis, the following environmental control strategies are formulated: 1. Acoustic environment optimization Applicable scenario: The user's sleep quality characteristics show frequent awakenings or fragmented sleep.

[0116] Specific measures: Noise detection: Monitor the nighttime noise level in the bedroom through wearable devices or environmental sensors.

[0117] Tiered intervention: If the noise level is >40 decibels: Install soundproof curtains, seal door and window gaps, or use active noise-canceling headphones.

[0118] If there is sudden noise (such as the sound of a door closing): introduce a white noise machine (such as the sound of a fan or rain) to cover up the sudden noise.

[0119] Effect verification: Monitor whether the number of awakenings is reduced after the intervention (target reduction ≥30%).

[0120] 2. Lighting control Applicable scenarios: Users who frequently get up at night or feel dizzy.

[0121] Specific measures: Lighting renovation: Install induction floor lights (brightness <10 lux, color temperature <2700K) to avoid strong light stimulation at night.

[0122] Remove the main bedroom light and use a dimmable bedside reading lamp instead.

[0123] Adjust furniture layout: Ensure that walking paths are clear at night to avoid tripping over debris.

[0124] Effect verification: After the intervention, the incidence of dizziness when getting up at night was counted (target reduction ≥50%).

[0125] 3. Temperature and humidity control Applicable scenario: The user's sleep quality characteristics show short deep sleep duration or frequent jitters during REM sleep.

[0126] Specific measures: Temperature Control: Summer: Set the air conditioner temperature to 24-26℃ to avoid sweating due to overheating or curling up due to overcooling.

[0127] Winter: Use an electric blanket (low setting) or thick bedding to maintain body surface temperature at 32-34°C.

[0128] Humidity regulation: If humidity is <30%: Use a humidifier to maintain humidity at 40-60% (to reduce respiratory dryness).

[0129] If humidity > 70%: Use a dehumidifier to prevent mold growth.

[0130] Effect verification: Monitor whether the deep sleep duration increases after the intervention (target increase ≥15%).

[0131] Based on sleep quality characteristics, the following exercise program is recommended to improve sleep and reduce the risk of falls: 1. Tai Chi training Applicable scenarios: Users with poor balance or frequent jitters during REM sleep.

[0132] Specific measures: Action selection: Give priority to movements that emphasize the transfer of center of gravity, such as "Cloud Hands", "Single Whip", and "Wild Horse's Mane".

[0133] Avoid standing on one leg, such as the "one-legged pose" (if the user has very poor balance).

[0134] Training plan: 3-5 times a week, 30-45 minutes each time, divided into 2-3 groups.

[0135] Combined with breathing exercises (such as "inhale when rising and exhale when falling"), it can enhance the autonomic nervous system regulation ability.

[0136] Effect verification: After the intervention, the improvement of balance ability was assessed through balance tests (such as the Berg scale).

[0137] 2. Balance and strength training Applicable scenarios: Users experience dizziness when getting up at night or lower limb muscle weakness.

[0138] Specific measures: Standing on one leg: Do 3 sets a day, 30 seconds each set (you can hold on to a wall for assistance), and gradually transition to not holding on to a wall.

[0139] Make it harder: Stand on one leg with your eyes closed or on a cushion.

[0140] Dynamic balance exercises: "Heel-toe walking": Walk in a straight line with your front heel touching your back toes, 10 meters at a time.

[0141] “Sit-to-Stand Test”: Stand up from a chair and sit down without support, recording the time (target < 10 seconds).

[0142] Lower limb strength training: Wall squats: 3 sets a day, 30 seconds each set, with knees not exceeding toes.

[0143] Elastic band resistance training: perform hip abduction, knee flexion and extension, etc.

[0144] Effect verification: The timed up-and-go test (TUGT) was used to assess the risk of falls after the intervention.

[0145] 3. Flexibility training Applicable scenarios: Users experience nocturnal muscle stiffness or limited body position changes.

[0146] Specific measures: Stretch before bed: Calf triceps: Stand facing a wall with your hands on the wall, straighten your back foot, and press your heel into the ground to feel the stretch on the back of your calf.

[0147] Quadriceps: Stand on one leg and grab your ankle with your other hand, stretching it toward your hip.

[0148] Back muscles: Lie on your back, bend your knees, wrap your hands around your knees and gently pull them towards your chest.

[0149] Training plan: Hold each movement for 15-30 seconds and repeat 2-3 times.

[0150] Combine with deep breathing (relax as you inhale, deepen the stretch as you exhale).

[0151] Effect verification: Flexibility improvement was assessed through joint range of motion testing after intervention.

[0152] Personalized intervention plan generation process: 1. Data input: Sleeping environment information (noise, light, temperature and humidity, and anti-slip properties).

[0153] Sleep quality characteristics (number of awakenings, REM jitters, AHI, and frequency of body position changes).

[0154] 2. Risk factor identification: Identify key risk factors through correlation analysis (such as Pearson correlation coefficient) or decision tree model.

[0155] Example: If the correlation coefficient between the number of awakenings and the noise level is > 0.5, then noise is the key factor.

[0156] 3. Strategy matching: Match environmental control strategies and exercise recommendations to risk factors.

[0157] For example: Scenario Example 1: Parkinson's Disease Patient Sleep quality characteristics: frequent limb shaking during REM phase (>2 times / minute) and frequent awakenings at night (>8 times / night).

[0158] Environmental information: Background noise in the bedroom is >45 decibels (traffic noise outside the window), there is no floor light at night, and you need to turn on the main light to go to the toilet.

[0159] Intervention options: Environment: Install soundproof windows + white noise machine, and install induction floor lights next to the bed.

[0160] Exercise: Daily Tai Chi training and calf stretching before bed.

[0161] Scenario Example 2: Diabetic Patients Sleep quality characteristics: Apnea-hypopnea index (AHI) > 15 times / hour, large blood sugar fluctuations during sleep (> 10%).

[0162] Environmental information: The bedroom's nighttime lighting is too bright (color temperature 4000K), miscellaneous items are piled near the bed, and the walking space is narrow.

[0163] Intervention options: Environment: Replace the light bulbs with warm lights and clear the clutter around the bed.

[0164] Exercise: Daily balance training (standing on one leg) + flexibility training (full body stretching).

[0165] Figure 2 is a module diagram of a fall warning system based on big data sleep quality analysis according to some embodiments of this specification, such as Figure 2As shown, the fall warning system for sleep quality analysis based on big data may include a big data analysis module, a model building module, a type determination module, a sleep monitoring module, a feature extraction module and a risk prediction module.

[0166] A big data analysis module is used to obtain big data samples, determine multiple user types, the sleep quality factors associated with each user type, and the weights of the sleep quality factors; A model building module is used to build a fall risk prediction model corresponding to each user type; A type determination module, used to determine the user type of the current user; Sleep monitoring module, used to obtain the sleep monitoring data of the current user; a feature extraction module, configured to extract a sleep quality feature of the current user from the sleep monitoring data of the current user based on a sleep quality factor associated with the user type of the current user; The risk prediction module is used to predict the fall risk of the current user based on the sleep quality characteristics of the current user through the fall risk prediction model corresponding to the user type of the current user.

[0167] The fall warning system based on big data sleep quality analysis is used to execute the fall warning method based on big data sleep quality analysis, which will not be described in detail here.

[0168] Finally, it should be understood that the embodiments described in this specification are intended only to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly described and illustrated in this specification.

Claims

1. A fall warning method based on big data sleep quality analysis, characterized in that: include: Obtaining a large data sample, determining multiple user types, sleep quality factors associated with each user type, and weights of the sleep quality factors; For each user type, a fall risk prediction model corresponding to the user type is established; Determine the user type of the current user; Obtain the current user's sleep monitoring data through the sleep monitoring device; Extracting a sleep quality feature of the current user from the sleep monitoring data of the current user based on a sleep quality factor associated with the user type of the current user; The fall risk prediction model corresponding to the current user's user type is used to predict the current user's fall risk based on the current user's sleep quality characteristics.

2. The fall warning method based on big data sleep quality analysis according to claim 1, characterized in that: The big data sample includes medical records, sleep monitoring data, and fall risk assessment data of multiple sample users: Identify multiple user types, including: Identify multiple diseases and drugs to be screened; Based on the medical records and fall data of multiple sample users, determine the impact coefficient of each disease to be screened on the fall risk, and screen the influencing diseases from the multiple diseases to be screened based on the impact coefficient of each disease to be screened on the fall risk; Based on the medical records and fall data of multiple sample users, determine the impact coefficient of each drug to be screened on the fall risk, and screen the influencing drugs from the multiple drugs to be screened based on the impact coefficient of each drug to be screened on the fall risk; Based on influencing diseases, influencing drugs and medical record information of multiple sample users, multiple user types are determined.

3. The fall warning method based on big data sleep quality analysis according to claim 2, characterized in that: Based on the medical history information of the affected diseases, affected drugs and multiple sample users, multiple user types are identified, including: For each sample user, determine the sample user's influencing disease identification vector based on the sample user's medical record information and influencing diseases; determine the sample user's influencing drug identification vector based on the sample user's medical record information and influencing drugs; Determine the similarity of the identification vectors of the two sample users based on the identification vectors of the affected diseases and the identification vectors of the affected drugs of any two sample users; Through the clustering algorithm, multiple sample users are divided into multiple user clusters according to the similarity of the identification vectors of any two sample users, and each user cluster corresponds to a user type.

4. The fall warning method based on big data sleep quality analysis according to any one of claims 1 to 3, characterized in that: Determine the sleep quality factors associated with each user type, including: Identify multiple sleep quality factors and fall risk assessment indicators to be screened; For each user type, based on the sleep monitoring data and fall risk assessment data of multiple sample users included in the user cluster, the correlation coefficient between each sleep quality factor to be screened and each fall risk assessment indicator is determined. According to the correlation coefficient between each sleep quality factor to be screened and each fall risk assessment indicator, the sleep quality factor associated with the user type is determined.

5. The fall warning method based on big data sleep quality analysis according to claim 4, characterized in that: The multiple sleep quality factors to be screened include at least the number of awakenings, total sleep time, first stage sleep time, and second stage sleep time; The multiple fall risk assessment indicators include at least vestibular nerve signal conduction efficiency, creatine kinase level, melatonin secretion cycle and systolic blood pressure.

6. The fall warning method based on big data sleep quality analysis according to claim 4, characterized in that: Determine the weight of the sleep quality factor associated with the user type, including: The weight of the sleep quality factor associated with the user type is determined according to the correlation coefficient between each sleep quality factor associated with the user type and each fall risk assessment indicator.

7. The fall warning method based on big data sleep quality analysis according to any one of claims 1 to 3, characterized in that: The sleep monitoring device includes a millimeter wave radar component and a heart rate monitoring component.

8. The fall warning method based on big data sleep quality analysis according to any one of claims 1 to 3, characterized in that: Based on the sleep quality factor associated with the user type of the current user, the sleep quality features of the current user are extracted from the sleep monitoring data of the current user, including: For each user type, a feature extraction model corresponding to the user type is established based on the sleep quality factor associated with the user type; The sleep quality features of the current user are extracted from the sleep monitoring data of the current user through a feature extraction model corresponding to the user type of the current user.

9. The fall warning method based on big data sleep quality analysis according to any one of claims 1 to 3, characterized in that: Also includes: If the current user's fall risk is greater than the preset fall risk threshold, the environment control strategy and exercise recommendations are determined based on the current user's sleeping environment information and the current user's sleep quality characteristics.

10. A fall warning system based on big data sleep quality analysis, characterized by: The fall warning method for sleep quality analysis based on big data according to any one of claims 1 to 9 comprises: A big data analysis module is used to obtain big data samples, determine multiple user types, the sleep quality factors associated with each user type, and the weights of the sleep quality factors; A model building module is used to build a fall risk prediction model corresponding to each user type; A type determination module, used to determine the user type of the current user; Sleep monitoring module, used to obtain the sleep monitoring data of the current user; a feature extraction module, configured to extract a sleep quality feature of the current user from the sleep monitoring data of the current user based on a sleep quality factor associated with the user type of the current user; The risk prediction module is used to predict the fall risk of the current user based on the sleep quality characteristics of the current user through the fall risk prediction model corresponding to the user type of the current user.

Images