A method for individual mortality prediction based on activity data of a wearable device
By acquiring multi-source feature data from wearable devices, calculating the average activity data feature value, and combining it with a nonlinear machine learning model, the problem of insufficient utilization of continuous, high-dimensional activity data in traditional models is solved, and more accurate individual mortality prediction is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN NOEN MEDICAL EQUIP CO LTD
- Filing Date
- 2026-02-25
- Publication Date
- 2026-07-10
AI Technical Summary
Existing mortality prediction models rely heavily on traditional data sources and lack effective integration and utilization of continuous, high-dimensional activity data, leading to inaccurate risk assessments.
By acquiring multi-source characteristic data of target individuals, including activity data and baseline characteristic data, and using wearable devices for continuous monitoring, the average characteristic value of the activity data is calculated, and then input into a nonlinear machine learning prediction model along with the baseline characteristic data to establish a nonlinear correlation between the individual's mortality risk and the risk of death.
It enables risk assessment to be updated in sync with an individual's daily activity level, captures key nonlinear patterns, and improves the accuracy and medical consistency of mortality prediction.
Smart Images
Figure CN122369903A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of mortality prediction technology, specifically to a method for predicting individual mortality rates based on activity data from wearable devices. Background Technology
[0002] Mortality prediction plays a crucial role in healthcare, and mortality prediction technologies can provide physicians with critical risk warnings and assist in clinical decision-making, thus gaining widespread attention in the medical field. However, many mortality prediction models rely heavily on traditional data sources (such as hospital records and demographic data), lacking the effective integration and utilization of continuous, high-dimensional activity data, which requires further improvement. Summary of the Invention
[0003] The embodiments of this application provide an individual mortality prediction method based on activity data from wearable devices, which can solve the technical problem that mortality prediction models rely heavily on traditional data sources and lack effective fusion and utilization of continuous, high-dimensional activity data.
[0004] In a first aspect, embodiments of this application provide a method for predicting individual mortality rates based on activity data from wearable devices, comprising the following steps: Data acquisition steps: Acquire multi-source feature data of the target individual. The multi-source feature data includes at least activity data and baseline feature data. The activity data is collected by the wearable device worn by the target individual during a continuous monitoring period. The baseline feature data includes at least one of demographic features, health features, living environment features, and lifestyle features. Feature construction steps: Process the activity data to calculate the average activity data feature value representing the overall activity level of the target individual; Model prediction steps: The average activity data feature values and the baseline feature data are input into a trained nonlinear machine learning prediction model, and the prediction model outputs the predicted mortality risk value of the target individual within a preset time window in the future.
[0005] In one embodiment, the average activity data characteristic value is the arithmetic mean of the activity data at all valid monitoring times within the continuous monitoring period; and / or The activity data includes at least one of heart rate, steps, distance traveled, cadence, acceleration, and speed.
[0006] In one embodiment, the activity data is acceleration data, and the feature construction step is: processing the acceleration data to calculate the average acceleration data feature value representing the overall activity level of the target individual; The prediction model is configured to characterize the nonlinear correlation between the mean acceleration data feature value and the risk of death. The nonlinear correlation is as follows: when the mean acceleration data feature value is below a first threshold, the risk of death is in a high plateau phase; when the mean acceleration data feature value is between the first threshold and a second threshold, the risk of death decreases significantly with the increase of the mean acceleration data feature value; when the mean acceleration data feature value is above the second threshold, the trend of the risk of death decreasing with the increase of the mean acceleration data feature value weakens or enters another low plateau phase.
[0007] In one embodiment, the nonlinear machine learning prediction model is an ensemble model based on a random forest model or a gradient boosting decision tree model.
[0008] In one embodiment, the health characteristics include at least one of the following: body mass index, body fat percentage, systolic blood pressure, diastolic blood pressure, history of diabetes, history of cancer, history of heart disease, recent hospitalization history, and long-term medication use; and / or The demographic characteristics include at least one of the following: age, sex, education level, income level, and marital status; and / or The lifestyle characteristics include at least one of the following: sleep duration, smoking status, and alcohol consumption status; and / or The living environment characteristics include at least one of the following: air quality, water quality, noise pollution level, climate conditions, and safety of the living environment.
[0009] In one embodiment, a model training step is further included for generating the trained nonlinear machine learning prediction model, the model training step comprising: Obtain historical population datasets, where each sample includes the individual's historical multi-source feature data and the corresponding actual mortality outcome label; Based on the historical population dataset, candidate nonlinear machine learning models are trained; The performance of the trained model is evaluated using a validation set, and the model with the best performance is identified as the trained nonlinear machine learning prediction model.
[0010] In one embodiment, during the model training step, a strategy for handling class imbalance is employed, wherein the strategy involves randomly downsampling the majority class samples.
[0011] Secondly, this application also provides an individual mortality prediction system based on wearable device data, comprising: The data acquisition module is configured to perform the above data acquisition steps; The feature construction module is configured to perform the feature construction steps described above; The model prediction module stores the trained nonlinear machine learning prediction model and is configured to perform the model prediction steps described above. The output module is configured to output the predicted mortality risk value.
[0012] Thirdly, this application also provides an electronic device, including: a memory and at least one processor; The memory contains computer programs; The at least one processor executes the computer program stored in the memory, causing the at least one processor to perform the individual mortality prediction method as described above.
[0013] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the individual mortality prediction method as described above.
[0014] The beneficial effects of the embodiments of the present invention are as follows: The present invention provides an individual mortality prediction method based on activity data from wearable devices. This method introduces raw activity data continuously and objectively monitored by wearable devices and extracts it into average activity data feature values. These average activity data feature values are a precise and unbiased measure of an individual's overall physical activity level. By integrating these average activity data feature values with traditional baseline feature data (such as demographic characteristics, health characteristics, living environment characteristics, and lifestyle characteristics), the model receives a richer and more reliable information source, enabling risk assessment to be updated synchronously with an individual's daily activity level. Furthermore, the nonlinear machine learning prediction model employed in this invention can automatically learn and characterize the specific nonlinear correlation between average activity data feature values and mortality risk, capturing key, easily overlooked nonlinear patterns and making the predictions more consistent with medical practice. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a flowchart of an embodiment of the present application of an individual mortality prediction method based on activity data from wearable devices; Figure 2 This is a flowchart illustrating the generation of a trained nonlinear machine learning prediction model according to one embodiment of this application; Figure 3This is a schematic diagram of the structure of an individual mortality prediction system based on wearable device data according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application; Figure label: 1. Data acquisition module; 2. Feature construction module; 3. Model prediction module; 4. Output module; 10. Memory; 20. Processor. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Furthermore, it should be understood that the specific embodiments described herein are only for illustration and explanation of the present invention and are not intended to limit the present invention. In the present invention, unless otherwise stated, directional terms such as "upper" and "lower" generally refer to the upper and lower positions of the device in actual use or operation, specifically the drawing directions in the accompanying drawings; while "inner" and "outer" refer to the outline of the device.
[0018] Please refer to Figure 1 This invention proposes a method for predicting individual mortality rates based on activity data from wearable devices, comprising the following steps: S1. Data acquisition steps: Acquire multi-source characteristic data of the target individual. The multi-source characteristic data includes at least activity data and baseline characteristic data. The activity data is collected by the wearable device worn by the target individual during a continuous monitoring period. The baseline characteristic data includes at least one of demographic characteristics, health characteristics, living environment characteristics, and lifestyle characteristics. S2. Feature construction steps: Process the activity data and calculate the average activity data feature value that represents the overall activity level of the target individual; S3. Model prediction steps: Input the average activity data feature values and baseline feature data into the trained nonlinear machine learning prediction model, and the prediction model outputs the predicted mortality risk value of the target individual within a preset time window in the future.
[0019] In this embodiment, compared to traditional models that heavily rely on self-reported lifestyle data (such as activity levels) or outdated medical records, which suffer from recall bias, strong subjectivity, and untimely updates, the individual mortality prediction method based on wearable device activity data of this invention introduces raw activity data continuously and objectively monitored by wearable devices and extracts it into average activity data feature values. These average activity data feature values are a precise and unbiased measure of an individual's overall physical activity level. By fusing these average activity data feature values with traditional baseline feature data (such as demographic characteristics, health characteristics, living environment characteristics, and lifestyle characteristics), the model is provided with a richer and more reliable information source, enabling risk assessment to be updated synchronously with an individual's daily activity level. Furthermore, the nonlinear machine learning prediction model used in this invention can automatically learn and characterize the specific nonlinear correlation between average activity data feature values and mortality risk, capturing key, easily overlooked nonlinear patterns and making the predictions more consistent with medical reality.
[0020] In one embodiment, the average activity data characteristic value is the arithmetic mean of activity data from all valid monitoring moments within a continuous monitoring period. This average reflects the average intensity of an individual's physical activity within the monitoring period, eliminating interference from instantaneous peaks or troughs. The activity data originates from raw activity data collected by wearable devices (e.g., smartwatches, smart bracelets, professional sports watches, smart glasses, sports chest straps, etc.) worn by the target individual over a continuous monitoring period (e.g., 7 consecutive days). Before calculation, the raw activity data needs to be calibrated and its validity assessed, discarding invalid data segments where the device was not worn, the signal was abnormal, or the monitoring duration was insufficient, ensuring that the analysis is based on data from valid monitoring moments. In this embodiment, the simplest arithmetic mean is used, providing the most basic but essential continuous activity intensity information for a powerful nonlinear model.
[0021] In one embodiment, the activity data includes at least one of heart rate, steps, distance traveled, cadence, acceleration, and speed.
[0022] In one embodiment, demographic characteristics include at least one of age, sex, education level, income level, and marital status.
[0023] In one embodiment, lifestyle characteristics include at least one of sleep duration, smoking status, and alcohol consumption status.
[0024] In one embodiment, health characteristics include at least one of body mass index, body fat percentage, systolic blood pressure, diastolic blood pressure, history of diabetes, history of cancer, history of heart disease, recent hospitalization, and long-term medication use.
[0025] In one embodiment, the living environment characteristics include at least one of air quality, water quality, noise pollution level, climate conditions, and living environment safety (such as local crime rate, presence of war, etc.).
[0026] In one embodiment, the activity data is acceleration data. The feature construction steps are as follows: processing the acceleration data to calculate the average acceleration data feature value representing the overall activity level of the target individual; the prediction model is configured to characterize the nonlinear correlation between the average acceleration data feature value and the mortality risk. The nonlinear correlation is as follows: when the average acceleration data feature value is below a first threshold, the mortality risk is in a high plateau phase; when the average acceleration data feature value is between the first and second thresholds, the mortality risk decreases significantly with the increase of the average acceleration data feature value; when the average acceleration data feature value is above the second threshold, the decreasing trend of mortality risk with the increase of the average acceleration data feature value weakens or enters another low plateau phase. In this embodiment, the mortality risk in the high plateau phase is greater than the mortality risk in the low plateau phase.
[0027] Traditional linear models or expert-based hierarchical models (such as simply categorizing activity levels as "low," "medium," and "high") systematically introduce prediction bias when processing acceleration data. For example, linear models may incorrectly assume that increasing activity levels from very low to moderate levels results in the same risk reduction as increasing from moderate to very high levels, which is clearly not physiologically accurate. This embodiment, however, inputs the average acceleration feature values into a specifically configured nonlinear machine learning prediction model. This model does not simply treat activity levels and risk as a linear negative correlation, but is trained to characterize a complex yet biologically sound nonlinear relationship between the two.
[0028] The nonlinear model in this invention automatically corrects for the biases of traditional linear models by capturing the three-stage relationship of "high plateau phase - steep decline phase - low plateau phase". Specifically, for a high-risk individual whose activity level is below a first threshold (e.g., the first threshold is 10 mg of average acceleration), the core message of health intervention is that "average acceleration needs to exceed the first threshold and enter the effective activity range to significantly reduce the risk of death." This provides a clear and quantifiable intervention target (e.g., suggesting that the target individual increase their average acceleration to above 10 mg). However, if a traditional linear model were used, it might only give a vague suggestion of "increasing activity". Similarly, for a low-risk individual whose activity level is above a second threshold (e.g., the second threshold is 40 mg of average acceleration), the risk prediction given by the model will be relatively stable, indicating that the health benefits of further vigorous activity are limited. This helps the target individual avoid excessive or unnecessary exercise. In addition, the model can integrate acceleration data with baseline feature data to quantitatively reveal the specific values of the first and second thresholds for different populations and individuals (for example, for a population aged 20-30, the first threshold is about 15mg and the second threshold is about 35mg).
[0029] In one embodiment, the nonlinear machine learning prediction model is an ensemble model based on a random forest model or a gradient boosting decision tree model.
[0030] In one embodiment, the individual mortality prediction method based on activity data from wearable devices further includes a model training step for generating the trained nonlinear machine learning prediction model, referencing... Figure 2 The model training steps include: S11. Obtain historical population datasets, where each sample includes the individual's historical multi-source feature data and the corresponding actual death outcome label; S12. Train candidate non-linear machine learning models based on historical population datasets; S13. Use the validation set to evaluate the performance of the trained model and identify the best-performing model as the trained nonlinear machine learning prediction model.
[0031] Step S11 requires obtaining a historical population dataset with a uniform structure, where each sample must contain historical multi-source feature data (such as formatted acceleration features and standardized baseline features) and authoritatively verified actual mortality outcome labels (such as those from the national death registry system). This step mandates the integration of raw data from different sources and formats (questionnaires, sensors, hospital records) into structured analytical units, ensuring that all subsequent models are learned and compared on the same data definitions and scales. For example, the average acceleration data for all samples are calculated using the same formula (arithmetic mean), and all recent hospitalization histories are defined within the same time window (e.g., within one year). This standardization of input is a fundamental prerequisite for obtaining stable, reliable, comparable, and reproducible predictive models across different studies, overcoming the problems of model performance fluctuations and unreliable conclusions caused by inconsistent data preprocessing in existing technologies.
[0032] Steps S12 and S13 together constitute a systematic model selection and evaluation framework. This framework does not pre-specify a single model (e.g., requiring logistic regression), but instead trains candidate nonlinear machine learning models (such as random forests, gradient boosting trees, neural networks, etc.) to impartially identify the nonlinear machine learning model best suited to the current data pattern and prediction task. The core of step S13 lies in the use of a validation set for evaluation. The validation set simulates the model's performance on unknown future individuals. Only models that demonstrate stable and excellent performance on the validation set are determined as the final usable trained nonlinear machine learning prediction model, ensuring that the final model not only fits historical data well but also makes accurate risk assessments for newly emerging and unseen individuals.
[0033] In one embodiment, during the model training step, a strategy for handling class imbalance is employed, which involves randomly downsampling the majority class samples.
[0034] The following uses a preferred embodiment of the present invention as an example to specifically illustrate how the final prediction model is obtained through the steps described: S11: Obtain historical population datasets S111: Obtain data from 10,000 participants from the UK Biobank and calculate the average acceleration (in mg) of each participant's 7-day wrist-worn accelerometer data.
[0035] S112: Extract baseline feature data for each participant (including: age, sex, body fat percentage, systolic blood pressure, history of diabetes (yes / no), history of cancer (yes / no), hospitalization history within one year (yes / no), smoking status, etc., for a total of 8 features).
[0036] S113: Obtain information on whether each participant died within three years after the acceleration monitoring ended from the linked national death registry system (yes=1, no=0). Combine each participant's feature vector (1 acceleration feature and 8 baseline features) with the corresponding death label (0 or 1) to form a structured historical population dataset containing 10,000 samples.
[0037] S12: Train the candidate model based on historical population datasets. S121: Data partitioning: Randomly divide the 10,000 samples into three parts: training set (7,000 people), validation set (1,500 people), and test set (1,500 people). The validation set is used for model selection in step S13, and the test set is used for final evaluation. The validation set and the test set do not participate in model selection and training at any point.
[0038] S122: Candidate Model Selection: Three representative nonlinear machine learning models were selected as candidates, including: Random Forest, Gradient Boosting Decision Tree, and Multilayer Perceptron Neural Network.
[0039] S123: Train each candidate model separately on the training set. For example, when training a random forest, use five-fold cross-validation to optimize its hyperparameters (such as the number of trees and the number of features considered in each split) on the training set. To handle the extreme imbalance of death labels (sparse death cases), randomly downsample the majority class (non-death) within each training fold of cross-validation to achieve a 1:4 ratio of dead to non-death samples. Each model uses the same preprocessing and the same training strategy.
[0040] S13: Use the validation set to evaluate and determine the optimal model. S131: Performance evaluation: Perform predictions on the validation set, which was not used for training, for the three trained candidate models.
[0041] S132: Evaluation Metric: Calculate the area under the receiver operating characteristic curve (AUC) for each model's predicted probability. AUC comprehensively measures the model's ability to distinguish between "dying" and "living" individuals and is the preferred evaluation metric for medical prediction models.
[0042] S133: Model Determination: Compare the AUC values of the three models on the validation set. Assume the results are: Random Forest (AUC = 0.780), Gradient Boosting Tree (AUC = 0.771), and Neural Network (AUC = 0.758). The Random Forest model performs best on the validation set.
[0043] S134: Model Determination: The random forest model and all its parameters that have been trained on the training set and validated on the validation set to achieve the best performance are formally determined as the trained nonlinear machine learning prediction model, ready for use in predicting new individuals.
[0044] Furthermore, this invention also provides an individual mortality prediction system based on wearable device data, such as... Figure 3 As shown, it includes: Data acquisition module 1 is configured to perform the data acquisition steps described above; Feature construction module 2 is configured to perform the feature construction steps described above; Model prediction module 3 stores a trained nonlinear machine learning prediction model and is configured to perform the model prediction steps as described above. Output module 4 is configured to output the predicted mortality risk value.
[0045] In addition, the present invention also provides an electronic device, such as Figure 4 As shown, it includes: a memory 10 and at least one processor 20; wherein, the memory 10 stores a computer program; the at least one processor 20 executes the computer program stored in the memory 10, causing the at least one processor 20 to perform the individual mortality prediction method as described above.
[0046] In some embodiments, memory 10 may be an internal storage unit of an electronic device, such as a hard disk or RAM. In other embodiments, memory 10 may be an external storage device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, or flash card. Furthermore, memory 10 may include both internal and external storage units. Memory 10 is used to store application software and various types of data installed on the electronic device, such as the program code of the electronic device.
[0047] In some embodiments, processor 20 may be a central processing unit (CPU), microprocessor 20 or other data processing chip, used to run program code stored in memory 10 or process data, such as executing an individual mortality prediction method based on activity data from wearable devices.
[0048] In addition, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the individual mortality prediction method as described above.
[0049] The following detailed embodiments further illustrate the above-mentioned method for predicting individual mortality rates based on activity data from wearable devices. Preferred embodiments of this application are described in detail below: Example
[0050] A mortality prediction method based on one-year average acceleration and a random forest model is presented in this embodiment. This embodiment demonstrates how to use acceleration data collected by an accelerometer, combined with baseline characteristics, to predict an individual's mortality risk over the next three years using a random forest model. The method includes the following steps: S10: Data Acquisition Target individual: Mr. Wang, male, 62 years old.
[0051] S101: Activity Data Acquisition: Mr. Wang has continuously worn an ST LIS2DW12 accelerometer for the past year. The device's built-in triaxial accelerometer collects raw acceleration signals at a frequency of 32Hz. All valid raw acceleration data from the past 365 consecutive days were obtained through the device's accompanying cloud service platform. The valid monitoring standard is a daily wearing time exceeding 10 hours.
[0052] S102: Baseline Characteristic Data Acquisition: Mr. Wang's baseline characteristic data was acquired through a health questionnaire and electronic health record system. The baseline characteristic data includes: Demographic characteristics: age (62 years old), gender (male), education level (undergraduate), annual income level (200,000-300,000 yuan).
[0053] Health characteristics: Body mass index (BMI, 26.5 kg / m²), systolic blood pressure (138 mmHg), diastolic blood pressure (85 mmHg), history of diabetes (none), history of cancer (none), history of heart disease (hypertension, no myocardial infarction or stroke), recent hospitalization history (none in the past year), long-term medication (daily antihypertensive medication).
[0054] Lifestyle characteristics: self-reported average daily sleep duration (6.5 hours), smoking status (quit smoking 10 years ago), and alcohol consumption status (occasional).
[0055] S20: Feature Construction S201: Data processing: For the acquired raw acceleration data of one year, sensor calibration and quality control are first performed to remove obvious outliers (such as continuous static for more than 24 hours is considered as the device not being worn).
[0056] S202: Calculate the average acceleration: For each valid monitoring moment (per second), calculate the composite vector amplitude of the triaxial acceleration. Then, take the arithmetic mean of the vector amplitudes of all valid monitoring moments over the past year to obtain the average acceleration characteristic value representing Mr. Wang's overall annual activity level. The calculated result is 28.5 mg.
[0057] S30: Model Prediction S301: Model Preparation: A trained random forest model is used. This model is trained on a historical dataset containing 10,000 people according to steps S11-S13 above, and shows the best performance on the independent validation set (AUC=0.78).
[0058] S302: Feature Input: Integrate Mr. Wang's average acceleration feature value (28.5 mg) and its baseline feature data into a feature vector and input it into the random forest model.
[0059] S303: Model Operation and Output: Hundreds of decision trees within the random forest model independently determine the feature vector. Each tree selects a path based on its internal splitting rules (such as "age > 60" or "average acceleration > 25mg") according to the feature value, ultimately deriving a death probability estimate at the leaf node. The probability estimates from all trees are averaged by voting to obtain the final ensemble prediction result.
[0060] S304: Risk Prediction Output: The model outputs a predicted mortality risk of 0.09 (i.e., a 9% probability) for Mr. Wang within the next three years. Based on the model's preset risk stratification thresholds (e.g., <5% is low risk, 5%-15% is medium risk, and >15% is high risk), Mr. Wang is assessed as medium risk. Example
[0061] A mortality prediction method based on three-year average heart rate and a gradient boosting decision tree model is presented in this embodiment. This embodiment demonstrates how to use heart rate data, combined with broader living environment characteristics, to predict mortality risk through a gradient boosting decision tree model. The method includes the following steps: S01: Data Acquisition Target individual: Ms. Li, female, 58 years old.
[0062] S011: Activity Data Acquisition: Ms. Li is a patient with stable chronic heart failure who has been wearing a medical-grade chest patch ECG monitor for the past three years. The device records her 24-hour ambulatory heart rate (HR) data daily. All valid resting heart rate data for the past 1095 consecutive days (three years) were acquired from the medical data center (defined as data collected during nighttime sleep).
[0063] S012: Baseline Characteristic Data Acquisition: Ms. Li's baseline characteristic data was obtained through a health questionnaire and electronic health record system. The baseline characteristic data includes: Demographic and health characteristics: age (58 years), sex (female), body mass index (24.0 kg / m²), history of heart disease (chronic heart failure, NYHA class II), recent hospitalization history (hospitalized two years ago due to acute exacerbation of heart failure), and long-term medication (taking beta-blockers and diuretics).
[0064] Living environment characteristics: obtained from environmental monitoring station data of the residence: annual average PM2.5 concentration (35 μg / m³), average nighttime noise level of the residence (45 dB), and average annual temperature of the residence (15°C). These characteristics were digitized and incorporated into the baseline feature set.
[0065] S02: Feature Construction S021: Data processing: Extract the effective daily resting heart rate data segments (usually specific time periods late at night) and remove outliers caused by arrhythmia or poor signal.
[0066] S022: Calculate the average heart rate: The arithmetic mean of the daily average resting heart rate over the past three years (approximately 1000 days) was calculated to obtain the average resting heart rate characteristic value representing Ms. Li's long-term cardiac resting load. The calculated result is 72 beats / minute.
[0067] S03: Model Prediction S031: Model Preparation: A trained gradient boosting decision tree (GBDT) model was used. This model was specifically trained on a cohort of cardiovascular disease patients, incorporating living environment features, and achieved an AUC of 0.82 on the validation set.
[0068] S032: Feature Input: Input Ms. Li’s mean resting heart rate feature value (72 bpm) and its baseline feature data into the GBDT model.
[0069] S033: Model Operation and Output: The GBDT model operates in a sequential manner. The first decision tree makes an initial prediction based on all features, and each subsequent tree is dedicated to correcting the residuals of the previous tree's predictions. After hundreds of such "boosting" iterations, the model integrates a powerful prediction function.
[0070] S034: Risk Prediction Output: The model outputs a predicted mortality risk of 0.18 (18%) for Ms. Li within the next year (for heart failure patients, short-term risk is often predicted). Combined with clinical thresholds, this is assessed as high risk.
[0071] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.
[0072] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.
[0073] The embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for predicting individual mortality rates based on activity data from wearable devices, characterized in that, Includes the following steps: Data acquisition steps: Acquire multi-source feature data of the target individual. The multi-source feature data includes at least activity data and baseline feature data. The activity data is collected by the wearable device worn by the target individual during a continuous monitoring period. The baseline feature data includes at least one of demographic features, health features, living environment features, and lifestyle features. Feature construction steps: Process the activity data to calculate the average activity data feature value representing the overall activity level of the target individual; Model prediction steps: The average activity data feature values and the baseline feature data are input into a trained nonlinear machine learning prediction model, and the prediction model outputs the predicted mortality risk value of the target individual within a preset time window in the future.
2. The individual mortality prediction method according to claim 1, characterized in that, The average activity data characteristic value is the arithmetic mean of the activity data at all valid monitoring times within the continuous monitoring period; and / or The activity data includes at least one of heart rate, steps, distance traveled, cadence, acceleration, and speed.
3. The individual mortality prediction method according to claim 2, characterized in that, The activity data is acceleration data, and the feature construction step is as follows: process the acceleration data and calculate the average acceleration data feature value that represents the overall activity level of the target individual; The prediction model is configured to characterize the nonlinear correlation between the mean acceleration data feature value and the risk of death. The nonlinear correlation is as follows: when the mean acceleration data feature value is below a first threshold, the risk of death is in a high plateau phase; when the mean acceleration data feature value is between the first threshold and a second threshold, the risk of death decreases significantly with the increase of the mean acceleration data feature value; when the mean acceleration data feature value is above the second threshold, the trend of the risk of death decreasing with the increase of the mean acceleration data feature value weakens or enters another low plateau phase.
4. The individual mortality prediction method according to claim 1, characterized in that, The nonlinear machine learning prediction model is an ensemble model based on random forest or gradient boosting decision tree models.
5. The individual mortality prediction method according to claim 1, characterized in that, The health characteristics include at least one of the following: body mass index, body fat percentage, systolic blood pressure, diastolic blood pressure, history of diabetes, history of cancer, history of heart disease, recent hospitalization, and long-term medication use; and / or The demographic characteristics include at least one of the following: age, sex, education level, income level, and marital status; and / or The lifestyle characteristics include at least one of the following: sleep duration, smoking status, and alcohol consumption status; and / or The living environment characteristics include at least one of the following: air quality, water quality, noise pollution level, climate conditions, and safety of the living environment.
6. The individual mortality prediction method according to claim 1, characterized in that, It also includes a model training step for generating the trained nonlinear machine learning prediction model, the model training step comprising: Obtain historical population datasets, where each sample includes the individual's historical multi-source feature data and the corresponding actual mortality outcome label; Based on the historical population dataset, candidate nonlinear machine learning models are trained; The performance of the trained model is evaluated using a validation set, and the model with the best performance is identified as the trained nonlinear machine learning prediction model.
7. The individual mortality prediction method according to claim 6, characterized in that, In the model training step, a strategy to handle class imbalance is adopted for training, which is to randomly downsample the majority class samples.
8. An individual mortality prediction system based on wearable device data, characterized in that, include: The data acquisition module is configured to perform the data acquisition steps as described in claim 1; A feature construction module is configured to perform the feature construction steps as described in claim 1; The model prediction module stores the trained nonlinear machine learning prediction model and is configured to perform the model prediction steps as described in any one of claims 1-7. The output module is configured to output the predicted mortality risk value.
9. An electronic device, characterized in that, include: Memory and at least one processor; The memory contains computer programs; The at least one processor executes the computer program stored in the memory, causing the at least one processor to perform the individual mortality prediction method as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the individual mortality prediction method as described in any one of claims 1 to 7.