A method and system for multi-dimensional biological age assessment using piecewise linear mapping

Abstract

The application discloses a kind of segmented linear mapping multi-dimensional biological age evaluation method and system, belong to wearable device data processing, health evaluation algorithm technical field.This scheme acquires five dimensions 10 physiological indexes by wearable device, adopts IQR+3σ two-level rule to clean abnormal data, relies on 7 days, 28 days, 365 days rolling window to build personal multi-cycle baseline;Combined with age, gender norm, the age offset of single index is quantified using segmented linear mapping, and the evaluation result can be completely traced back.The application adopts double time window single engine architecture, outputs biological age every week, and outputs aging speed every day;Coupled with missing weight dynamic distribution, ±0.5 year fluctuation limit speed, three optimization mechanisms of mutual exclusion of cardiopulmonary index, and hierarchical index automatically generates health intervention suggestions, solve seven big defects such as existing model uninterpretable, weak anti-interference, single evaluation dimension.

Description

Technical Field

[0001] This invention belongs to the field of wearable device data processing and health assessment algorithm technology, specifically relating to a method and system for dynamically assessing biological age and aging rate based on multidimensional physiological data and a piecewise linear mapping function. This invention relies on multidimensional health data collected by consumer-grade wearable devices, combined with a piecewise linear age influence mapping function and a dual-time-window single-engine architecture, to achieve a transparent, attributable, robust, and actionable two-layer dynamic assessment of biological age. Background Technology

[0002] Biological age is a comprehensive health indicator that reflects an individual's physiological aging relative to chronological age. When biological age is lower than chronological age, it indicates that the individual's physiological function is better than the average level of their peers; when biological age is higher than chronological age, it indicates that the individual is at risk of accelerated physiological decline. In recent years, consumer-grade wearable health devices have rapidly become widespread, continuously collecting multi-dimensional physiological data such as heart rate, sleep, and activity levels, providing a reliable data foundation for dynamic assessment of biological age.

[0003] Currently, mainstream bio-age assessment methods can be divided into three categories based on their technical principles. Each category has significant technical shortcomings, as detailed below:

[0004] The first category is a black-box joint evaluation scheme based on deep learning. This scheme uses deep learning architectures such as MSE-Net and Bi-LSTM, taking static features and dynamic physiological indicators as inputs, and outputting biological age through neural network regression. The aging rate is calculated by simply differencing the difference between biological ages at adjacent time points by the time interval. This type of scheme can introduce an attention mechanism to sort the output indicators by relative weights, but the algorithm as a whole is a black-box nonlinear regression, which cannot quantify the specific impact of a single indicator on biological age by years. At the same time, the calculation of the aging rate depends entirely on the absolute value of biological age, which is very easy to amplify data noise.

[0005] The second category is simple mapping schemes based on a single or few indicators. These schemes only select individual indicators such as maximum oxygen uptake, steps, and resting heart rate to calculate biological age bias. They have a single evaluation dimension and ignore the synergistic effects of multiple physiological systems such as sleep, physical recovery, and body composition, resulting in poor adaptability to individuals with different physical conditions.

[0006] The third category is a heuristic, manually weighted summation scheme based on multiple indicators. This scheme manually weights the scores of multiple health dimensions to calculate the biological age bias; however, the indicator weights are set based on human experience and have not been calibrated using norms from authoritative literature, resulting in weak statistical support. This type of scheme is susceptible to contamination by extreme data, lacks a fair handling mechanism for missing data, does not have anti-jump logic, only outputs results for a single time dimension, lacks independent aging rate calculation logic, and cannot provide users with actionable improvement suggestions.

[0007] Comprehensive analysis reveals seven common technical defects in existing biological age assessment technologies, and each defect leads to corresponding technical problems: (1) Uninterpretable assessment results: Deep learning solutions cannot quantify the contribution of a single indicator, and users cannot carry out targeted health improvement behaviors; (2) Weak anti-interference ability: Instantaneous values ​​or short-term averages are generally used as inputs, and a single extreme data can pollute the calculation results. The monthly fluctuation of the assessment results can exceed 20 years, losing reference value; (3) Unreasonable handling of missing data: There is no fair weight redistribution mechanism after the indicators are missing, which can easily cause systematic assessment bias; (4) Single assessment dimension: Only the current biological age is output, which cannot simultaneously reflect the long-term physiological state and recent behavioral change trends; (5) Low accuracy in aging speed calculation: Traditional differential calculation methods rely on the absolute value of biological age, resulting in serious noise amplification and unstable calculation results; (6) No practical guidance capability: Only the assessment results are output, which cannot be transformed into health improvement actions that users can directly implement; (7) Collinear features are not removed: Indicators with causal collinear relationships are included in the calculation at the same time, causing the corresponding dimension weights to be implicitly doubled, resulting in assessment distortion. Summary of the Invention

[0008] This invention relates to the following definitions of professional abbreviations and self-coined terms: (1) VO2max: maximum oxygen uptake, a core indicator characterizing human cardiopulmonary function; (2) Nocturnal heart rate variability: fluctuations in the intervals between consecutive heartbeats during the human sleep stage, used to assess autonomic nervous system function during sleep, abbreviated as nocturnal HRV; (3) RMSSD: root mean square of the difference between adjacent heartbeat intervals, a core indicator for analyzing nocturnal HRV; (4) IQR: interquartile range, a commonly used statistic for outlier determination; (5) 3σ criterion: rule for eliminating three times the standard deviation constant; (6) Sleep debt duration: the cumulative difference between an individual's actual sleep duration and their standard sleep requirement; (7) Oversleep guard: used to quantify the extent to which an individual's sleep duration exceeds their standard sleep requirement; (8) Sleep regularity score (SRI): a sleep stability score calculated based on fluctuations in sleep onset and wake-up times, with a score range of 0~100. (9) Vitality recovery: an indicator that represents the level of overnight recovery of bodily functions; (10) Controllable leverage indicator: a health indicator that users can directly adjust through daily behavior; (11) Outcome indicator: an indicator that reflects long-term physiological state and cannot be directly controlled by users; (12) Dual time window single engine architecture: the same segmented linear mapping calculation engine, accesses two sets of data from different time windows, and outputs two types of evaluation results in parallel; (13) Improved z-score: a standardized score used to quantify the short-term relative change of the indicator to the long-term baseline.

[0009] The core technical problem this invention aims to solve is to overcome the shortcomings of existing technologies, such as lack of interpretability, poor noise resistance, improper handling of missing values, coarse calculation of aging rate, and distorted weights of collinear indicators. Based on multidimensional health data collected from consumer-grade wearable devices, it constructs a transparent, attributable, and data-judgment-resistant biological age quantification model. Simultaneously, it designs independently computed aging rate indicators, ensuring that the two indicators share a computing engine, are independent of each other, and do not involve duplicate counting. Secondary optimization issues include: converting assessment results into actionable health actions to provide users with clear behavioral guidance; and supplementing the algorithm's full-boundary fallback logic to ensure stable operation under scenarios with low data volume, new users, and insufficient baseline.

[0010] To achieve the above objectives, the first aspect of this invention provides a multidimensional dynamic assessment method for biological age based on piecewise linear age impact mapping. This method relies on multidimensional health data collected by wearable devices, combining multi-period dynamic baselines, piecewise linear mapping, and a dual-time-window single-engine architecture to output biological age and aging rate in parallel, while simultaneously completing data attribution and action guidance. The method includes the following steps:

[0011] Step S1: Data Collection and Preprocessing. Data on 10 health indicators across five dimensions—sleep, recovery, physical fitness, activity, and body composition—is collected. Data validity is verified and extreme values ​​are removed. The 10 health indicators are: sleep debt duration, oversleep guard, sleep regularity score, energy recovery, VO2 max, resting heart rate, nocturnal heart rate variability, total daily steps, high-intensity exercise duration, and body mass index (BMI). VO2 max and high-intensity exercise duration are collinear indicators, and a mutually exclusive calculation rule is set: VO2 max data is prioritized; if the number of valid VO2 max data days is <14 days or the equipment estimation error is >10%, it is considered invalid, and high-intensity exercise duration is automatically switched to. The two indicators are not included in the calculation simultaneously.

[0012] This step employs a combined outlier removal strategy: IQR followed by 3σ. The first step, IQR removal, identifies data exceeding the range [Q1−1.5×IQR, Q3+1.5×IQR] as mild extrema and removes them. The second step, 3σ removal, identifies remaining data exceeding three times the standard deviation of [μ−3σ, μ+3σ] as extreme extrema and removes them. Overlapping extrema are uniformly removed. Indicator validity criteria: If the number of valid data points for a single indicator does not reach a preset threshold after two levels of extrema removal, the indicator is considered invalid. Only valid indicators participate in subsequent calculations; records with no valid data for a single day are left blank and not filled using interpolation.

[0013] Step S2: Multi-period dynamic baseline calculation. Three types of daily rolling time windows are set: 7-day, 28-day, and 365-day time windows. The median and standard deviation of each health indicator are calculated one by one to construct the user's personal historical baseline. Window usage rules: The median of the 365-day and 28-day time windows are used as long-window inputs for calculating biological age; the data change signals of the 28-day and 7-day time windows are used as short-window inputs for calculating the aging rate. Window design rationale: The 365-day long window represents the user's long-term physiological homeostasis, the 28-day medium window takes into account mid-term state changes, and the 7-day short window accurately captures short-term behavioral fluctuations.

[0014] Baseline downgrade rule: If the number of valid data days within a 365-day time window is less than 90 days, the 365-day baseline is abandoned, and a 28-day time window baseline is used uniformly in the calculation. The final result is marked "Calibrating," and the confidence interval of the result is expanded. Last-resort downgrade logic: If the number of valid data days within the 28-day window still does not meet the minimum number of valid data points, it is marked "Data is severely insufficient; results are for reference only," and the confidence interval is expanded a second time.

[0015] Personal sleep needs learning rules: Personal sleep needs are learned from the median sleep duration over a 90-day rolling history. The sleep duration is calculated as the total time from falling asleep in bed to waking up. New users will have a default standard sleep duration of 7.5 hours during the transition period, which is fixed at 90 days. After the transition period, the system will automatically switch to the personal learning baseline.

[0016] Step S3 involves calculating the piecewise linear age impact mapping. A piecewise linear mapping function is independently configured for each health indicator, using the individual's historical median baseline as input and the median reference value for age-gender stratified populations published in authoritative literature as a benchmark. Combining directional coefficients, slopes, and upper and lower limit rules, the indicator deviation is quantified into an age impact value. The age impact values ​​of all indicators can be independently extracted and linearly summed, achieving transparent attribution across all dimensions.

[0017] Further, the general formula of the piecewise linear age influence mapping function in step S3 is: AgeImpact_i= clamp( slope_i × dir_i × (value_i − ref_i), floor_i, cap_i ) + c_i, wherein the functions and parameters are defined as follows: clamp( X, L, U) is a limiting function; the operation rule is: output L when (X<L); output U when (X>U); output X when L<X < U; wherein L= floor_i is the lower limit of reverse aging, U=cap_i is the upper limit of additional aging; value_i: the median of personal long-term baseline of the i-th indicator; ref_i: the age × gender stratified authoritative reference value corresponding to the i-th indicator; dir_i: direction coefficient, which only takes +1 or -1, used to distinguish between two scenarios: "higher indicator value means worse condition" and "lower indicator value means worse condition"; slope_i: slope, representing the age change amplitude (years / unit) corresponding to each unit deviation of the indicator from the reference value; floor_i: the lower limit of reverse aging (minimum age influence value) corresponding to a single indicator; cap_i: the upper limit of additional aging (maximum age influence value) corresponding to a single indicator; c_i: neutral zero-point calibration constant, which is defaulted to 0 in the present invention and can be fitted and calibrated according to the P50 median of the target population.

[0018] Collinear supplementary processing rule for homologous indicators: In addition to the mutual exclusion of choosing one from two in the cardiopulmonary dimension, the correlation coefficient of homologous indicators including sleep and recovery indicators shall be calculated, and the weight of corresponding indicators shall be automatically attenuated when the correlation coefficient is greater than 0.7, so as to avoid the implicit superposition distortion of the weight of homologous indicators.

[0019] Step S4: Biological age slow-layer output: aggregate the age influence values of all effective indicators after weight adjustment, and superpose the user's actual chronological age to obtain the initial biological age; an output layer speed limit rule is set to suppress result jump, and a biological age result is output regularly once a week.

[0020] Missing indicator weight redistribution rules: W_total = Σ_all_i (cap_i − floor_i), which is the sum of the ranges of all 10 indicators, approximately 29.3 years; W_valid = Σ_valid_i (cap_i − floor_i), which is the sum of the ranges of the currently valid indicators; the formula for calculating the age impact after adjustment for each valid indicator k is: AgeImpact_k_adjusted = AgeImpact_k × (W_total / W_valid). When W_valid < W_total / 2 (more than half of the indicators are missing), the system degrades to "partial estimation" mode, expands the confidence interval in the output, and provides a prompt. The output is labeled "based on N / 10 indicators" (N is the number of valid indicators). If N < 4, the output is rejected, and the prompt "Insufficient data, please synchronize more health data and check" is displayed.

[0021] Weekly rate limit rule: Let Δ be the difference between the initial biological age calculated this week and the output biological age last week. Limit Δ to the range [-0.5, +0.5]. Final output biological age = last week's biological age + clamp(Δ, -0.5, +0.5). The rate limit threshold is set based on the physiological aging process of the human body and user experience, with a maximum weekly fluctuation limit of ±0.5 years, decoupling output stability from the quality of the original data.

[0022] Cold start initialization rules: For new users with no previous weekly biological age, only the baseline is continuously calculated for the first 4 weeks, and the biological age is not output; after 4 weeks, the initial smoothed biological age is output for the first time, and the weekly rate limit rule is enabled thereafter.

[0023] Step S5 outputs the aging speed layer. This step and step S4 are parallel computation branches, sharing the same piecewise linear mapping calculation engine, but their input data and calculation logic are independent, with no duplicate counting. An improvement z-score is calculated for the controllable leverage indicator, and a weighted improvement index is obtained, ultimately mapped to the aging speed multiplier. The output is updated daily.

[0024] Improve the z-score calculation rules: For conventional controllable leverage indicators, calculate using a 28-day baseline and a 365-day baseline; for dormant debt duration, a short-term highly sensitive indicator, calculate using a 7-day baseline and a 28-day baseline to improve the sensitivity of short-term behavior identification. After unifying the improvement direction, calculate the weighted average using the range of each indicator as the weight to obtain the improvement index I.

[0025] Aging rate mapping formula: Aging rate = clamp(1.0 - k × I, -1.0, 3.0) where the sensitivity coefficient k is fixed at 1.0, but can be adjusted according to user needs; the aging rate range [-1.0, 3.0] is limited by the physiological range of human aging rate. Result interpretation: Improvement index I>0 indicates recent behavioral improvement, aging rate less than 1.0, and slowed aging; I=0 indicates stable condition, aging rate equal to 1.0, in sync with natural aging; I<0 indicates recent regression, aging rate greater than 1.0, and accelerated aging.

[0026] Step S6 Actionable attribution display: The 10 health indicators are divided into two categories: controllable leverage indicators and outcome indicators. (1) Controllable leverage indicators: sleep debt duration, oversleep guard, sleep regularity score, total daily steps, high-intensity exercise duration, and body mass index; based on the age impact value of the indicators, the actionable behavior improvement actions that users can directly perform are output. (2) Outcome indicators: vitality recovery, maximum oxygen uptake, resting heart rate, and nocturnal heart rate variability (RMSSD); these indicators cannot be directly controlled, and the attribution results are forcibly redirected to the corresponding upstream controllable leverage indicators. The mapping relationship between outcome indicators and upstream controllable leverage indicators is shown in Table 1.

[0027] The fallback attribution logic is as follows: If all controllable sleep-related indicators are missing, vitality recovery and nighttime HRV are redirected to controllable indicators such as exercise and BMI, eliminating the failure gap in the attribution module. The mapping relationship between outcome-type indicators and upstream controllable leverage indicators is as follows: vitality recovery is redirected to sleep debt duration and sleep pattern score; VO2 max is redirected to high-intensity exercise duration and total daily steps; resting heart rate is redirected to high-intensity exercise duration and sleep debt duration; nighttime HRV (RMSSD) is redirected to sleep pattern score and high-intensity exercise duration.

[0028] Supplementary explanation of the asymmetric U-shaped curve of sleep debt and oversleep guard: The slope of the left side of the sleep debt curve is 1.5, with a range of 5.0 years; the slope of the right side of the oversleep guard curve is 0.3, with a range of 1.0 years; the difference in slope reflects the objective epidemiological law that "the physiological harm of insufficient sleep is far greater than that of excessive sleep".

[0029] The second aspect of this invention provides a multidimensional biological age dynamic assessment system based on piecewise linear age influence mapping, used to implement the aforementioned multidimensional biological age dynamic assessment method. The system adopts a dual-time-window single-engine architecture, with data exchange and time-series linkage among modules. The system includes: a data acquisition and preprocessing module, a multi-period dynamic baseline module, a piecewise linear age influence calculation engine module, a biological age slow layer output module, an aging speed fast layer output module, a missing value fair handling and credibility labeling module, an actionable attribution display module, and a personalized benchmarking configuration module.

[0030] Data acquisition and preprocessing module: used to collect raw data of 10 multidimensional health indicators, perform 3σ+IQR two-level extreme value removal and data validity verification, and output cleaned valid data; compatible with multi-brand consumer wearable devices, unified data synchronization cycle and sampling frequency, and blank data is retained for missing single-day data without interpolation.

[0031] Multi-period dynamic baseline module: Connects to the data acquisition and preprocessing module, calculates the median and standard deviation of each indicator based on daily rolling time windows of 7 days, 28 days, and 365 days, and generates personal historical baselines; at the same time, it executes baseline downgrade judgment logic.

[0032] Piecewise linear age impact calculation engine module: The core calculation unit of the system, with two independent read and write caches built in long window and short window; it receives output data from the multi-period dynamic baseline module, calls the piecewise linear mapping function to calculate the age impact value for each indicator; it provides unified calculation logic for the two output branches of biological age and aging rate, with data isolation and no cross-reuse.

[0033] Biological Age Slow Layer Output Module: Receives the output data from the computation engine, combines the missing value weight redistribution rule and the weekly change rate limit rule, and generates and outputs the biological age result weekly.

[0034] Fast Aging Layer Output Module: Works in parallel with the slow biological age layer output module, receives data from the computing engine, calculates the improvement z-score, improvement index, and aging speed multiplier, and generates and outputs aging speed results daily.

[0035] Missing value fair handling and credibility labeling module: Embedded inside the biological age slow layer output module, it completes the redistribution of missing indicator range weights, and calculates and labels the credibility of the results.

[0036] Actionable Attribution Display Module: Receives the calculation results of biological age and aging rate, outputs actionable actions or retargeting attribution suggestions according to the indicator type, completes the front-end visualization display, and has a built-in fallback attribution branch for missing indicators.

[0037] Personalized benchmarking configuration module: Connects to the multi-cycle dynamic baseline module to realize functions such as learning personal sleep needs, storing personal recovery baselines, and automatically matching age / gender stratified reference values.

[0038] Compared with existing technologies, this invention has the following core technical advantages, addressing the seven common defects of existing technologies one by one: (1) Regarding the defect of "uninterpretable evaluation results": This invention achieves completely transparent and attributable results. The age influence value of each indicator can be independently quantified and linearly summed, accurately displaying the age contribution of a single indicator, solving the problem that deep learning black box solutions cannot explain and users cannot carry out targeted health improvement behaviors; (2) Regarding the defect of "weak anti-interference ability": This invention uses a multi-period median baseline to suppress extreme value interference and superimposes a ±0.5 years / week rate limit rule in the output layer, solving the problem of single extreme data pollution and large jumps in evaluation results from both the data input and result output layers, ensuring stable and reliable evaluation results; (3) Regarding the defect of "unreasonable handling of missing data": This invention dynamically redistributes weights based on the indicator range ratio and combines it with credibility labeling to achieve fair handling of missing data, solving the problem that traditional solutions handle missing values ​​roughly and are prone to causing systematic evaluation bias; (4) Regarding the defect of "single evaluation dimension": The defects of the present invention are as follows: (5) Regarding the defect of "low accuracy of aging speed calculation": The aging speed of the present invention is based on the independent calculation of behavior improvement signal, which does not depend on the absolute value of biological age, thus avoiding the problems of noise amplification and unstable results of the traditional differential calculation method, while maintaining the sensitivity of short-term changes; (6) Regarding the defect of "no practical guidance capability": The present invention constructs a hierarchical attribution logic of controllable leverage indicators and result indicators, which can directly transform the evaluation results into health improvement actions that users can perform, thus solving the problem that the traditional solution only outputs the evaluation results and cannot be transformed into health improvement actions that users can directly perform; (7) Regarding the defect of "collinear features not removed": The present invention sets a two-choice mutual exclusion rule for cardiopulmonary indicators with causal collinearity, thus avoiding the problem of implicit doubling of corresponding dimension weights and evaluation distortion, and ensuring that the results are objective and accurate. Attached Figure Description

[0039] Figure 1 is a schematic diagram of the overall system architecture of the present invention; The diagram shows the overall layered structure of the system, which consists of a health data acquisition layer, a data preprocessing module, a multi-period dynamic baseline module, a piecewise linear age impact calculation engine, a dual-output module, and an attribution display module in sequence; Explanation of reference numerals: 101 - Health data acquisition layer, 102 - Data preprocessing module, 103 - Multi-period dynamic baseline module, 104 - Piecewise linear age impact calculation engine, 105 - Slow biological age layer output module, 106 - Fast aging layer output module, 107 - Missing value fair processing module, 108 - Actionable attribution display module, 109 - Personalized benchmarking configuration module.

[0040] Figure 2 is a schematic diagram of the complete calculation process of the present invention; it is used to illustrate the full process logic and branch judgment conditions of data from acquisition, preprocessing, baseline calculation, mapping operation to dual result output and attribution display.

[0041] Figure 3 is a schematic diagram of the range distribution of the age-related influence of the indicators of the present invention; the lower limit of anti-aging, the upper limit of aging and the range of each of the 10 indicators are shown in the form of a bar chart, and the mutually exclusive relationship between maximum oxygen uptake and high-intensity exercise duration is also marked.

[0042] Figure 4 is a schematic diagram of the asymmetric U-shaped age-related curve of sleep according to the present invention; the horizontal axis represents the sleep duration deviation (unit: h), and the vertical axis represents the age-related value (unit: years), showing the slope and boundary of the two curves for sleep debt and oversleep protection, respectively.

[0043] Figure 5 is a schematic diagram of the U-shaped age influence curve of BMI segmentation according to the present invention; the horizontal axis is the body mass index (BMI), and the vertical axis is the age influence value (unit: years), showing the age influence pattern corresponding to different BMI intervals. Detailed Implementation

[0044] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0045] Example 1: System Architecture Implementation (corresponding to) Figure 1 )

[0046] like Figure 1As shown, the system of this invention includes: a health data acquisition layer 101 that collects raw data on sleep, heart rate, exercise, and body composition through consumer-grade wearable devices, and transmits it to a data preprocessing module 102 to complete extreme value removal and validity verification; after cleaning, the data enters a multi-period dynamic baseline module 103 to calculate the median and standard deviation of three time windows. All baseline data are uniformly sent to a piecewise linear age impact calculation engine 104, which has two parallel outputs: a long-window data-driven slow-layer biological age output module 105, in conjunction with a missing value fair processing module 107, outputs biological age weekly; a short-window data-driven fast-aging layer output module 106 outputs aging rate daily. A personalized benchmarking configuration module 109 provides personalized parameters such as individual sleep needs and population reference values ​​to the multi-period dynamic baseline module 103; the two outputs are finally summarized to an actionable attribution display module 108 to complete the action suggestion output.

[0047] Example 2: Implementation of the complete calculation process (corresponding to) Figure 2 )

[0048] like Figure 2 As shown, the calculation steps are as follows: Step 201: Enter the user's actual calendar age and gender, and collect raw data for 10 health indicators; Step 202: Execute the 3σ+IQR joint strategy to remove extreme value data and complete preprocessing; Step 203: Determine if the number of valid indicators is ≥4; if less than 4, proceed to Step 204, output a "data insufficient" prompt, and terminate the calculation; Step 204: Output a data insufficient prompt; Step 205: Calculate the median and standard deviation of all indicators based on 7-day, 28-day, and 365-day daily rolling windows; Step 206: Determine if the number of valid data days in the 365-day window is ≥90 days; Step 207: If the data is less than 90 days, downgrade to using the 28-day baseline, mark "calibrating," and expand the confidence interval; Step 208: If the data meets the requirements, use the 365-day baseline normally; Step 209: Call the piecewise linear mapping general formula to calculate the age influence value for each indicator; Step 210: Perform range weighting reassignment for missing indicators and label the reliability of the results; Step 211: Summarize all adjusted age impact values; Step 212: Overlay actual age, execute weekly rate-limiting rules, and output biological age weekly; Step 213: Parallel branch: Calculate the improvement z-score for controllable leverage indicators; Step 214: Calculate the improvement index I using range weighting; Step 215: Map and calculate the aging rate, updating the output daily; Step 216: Differentiate indicator types and complete the actionable attribution display.

[0049] Example 3: Implementation of the effect of age on range distribution (corresponding to) Figure 3 )

[0050] like Figure 3 As shown, the lower limit of anti-aging and the upper limit of aging for each of the 10 indicators are displayed in a two-way horizontal bar chart. The sleep dimension (indicators 1-3) has a combined range of 9.0 years, accounting for 30.7% of the total range of 29.3 years, reflecting the design principle of sleep as the dominant dimension. Among them, the sleep debt duration range is 5.0 years (lower limit of anti-aging -1.0, upper limit of aging +4.0), which is the single indicator with the largest range; the oversleep guard range is 1.0 years; and the sleep regularity score range is 3.0 years. In the physical fitness dimension, the VO2max range is 5.0 years (lower limit of anti-aging -2.5, upper limit of aging +2.5), which is one of the single indicators with the largest range, but it is mutually exclusive with the high-intensity exercise duration (range 3.0 years), and is marked with a dashed box "choose one". The ranges of the other indicators are: vitality recovery 2.5 years, resting heart rate 3.0 years, nighttime HRV 2.5 years, total daily steps 2.0 years, and BMI 2.3 years.

[0051] Example 4: Implementation of the Asymmetric U-Shaped Age-Influence Curve on Sleep (corresponding to) Figure 4 )

[0052] like Figure 4 As shown, the horizontal axis represents the deviation in sleep duration (hours), and the vertical axis represents the impact of age (years). The left arm is the sleep debt curve with a slope of 1.5. When the accumulated debt d=0, it corresponds to the upper limit of age reduction of -1.0 year; when d=1.0h, it corresponds to +0.5 years; when d=2.0h, it corresponds to +2.0 years; and when d≥3.3h, it reaches the upper limit of age increase of +4.0 years. The right arm is the oversleep protection curve with a slope of 0.3. When oversleep amount over=0, it corresponds to 0 years; when over=1h, it corresponds to +0.3 years; and when over≥3.3h, it reaches the upper limit of age increase of +1.0 year. The difference in slope between the two arms (1.5 vs 0.3) accurately reflects the physiological priority that "the harm of insufficient sleep is far greater than that of oversleep," forming an asymmetric U-shaped curve.

[0053] Example 5: Implementation of the BMI-segmented U-shaped age effect curve (corresponding to) Figure 5 )

[0054] like Figure 5 As shown, the horizontal axis represents BMI value, and the vertical axis represents the impact of age (in years). A BMI in the range [18.5, 25) indicates a slight age-reversal zone, with the optimal point being a BMI of 22 (<50 years) or a BMI of 23 (≥50 years), close to the lower limit of age reversal by -0.3 years. A BMI in the range [25, 30) indicates overweight and aging, increasing linearly from +0.3 years. A BMI ≥30 reaches the upper limit of aging by +2.0 years. A BMI <18.5 indicates excessively low aging, with greater aging occurring the further away from the threshold. The overall relationship exhibits a segmented U-shaped pattern.

[0055] Example 6: Standardization of Ten Health Indicator Parameters

[0056] The classification, properties, and age-related range parameters of the 10 health indicators involved in this invention are shown in Table 2:

[0057] The total age range is 29.3 years; the sleep dimension indicators 1-3 combined range is 9.0 years, accounting for 30.7% of the total age range; indicator 5 (VO2max) and indicator 9 (duration of high-intensity exercise) are mutually exclusive and cannot be included in the calculation at the same time.

[0058] Example 6: Specific mapping formulas for each indicator

[0059] Sleep debt duration: Based on an individual's 90-day learning sleep needs as a zero-debt baseline, input the 14-day index-weighted sleep deficit median d(h) over 365 days. Formula: AgeImpact_debt = clamp(1.5 × d − 1.0, −1.0, +4.0). Parameter explanation: 1.5 is the epidemiological dose-response slope; -1.0 is a constant term in the formula, used to achieve age-reversal rewards for debt-free users; clamp constraint interval [-1.0, 4.0]. Validation points: d=0 → -1.0 years; d=1.8 → +1.7 years; d≥3.3 → capped at +4.0 years.

[0060] Oversleep Guardian: Oversleep tolerance is 1 hour, over = max(0, median sleep - individual sleep requirement - 1.0). Formula: AgeImpact_oversleep = min(+1.0, 0.3 × over). Checkpoints: over=0 → 0 years old; over=1 → +0.3 years old; over≥3.3 → capped at +1.0 years old.

[0061] Sleep regularity score (SRI): The population reference norm is 85 points, and the median SRI over 365 days is entered as s. Formula: AgeImpact_SRI = clamp(−0.5 + (85 − s) / 85 × 3.0, −0.5, +2.5). Checkpoints: s≥85 → lock in -0.5 years; s≤0 → lock in +2.5 years.

[0062] Recovery Performance: ref_r is the user's median recovery over 365 days (personal baseline), and r is the current median recovery over 28 days. Formula: AgeImpact_recovery = clamp(−0.5 + (ref_r − r) / ref_r × 2.5, −0.5, +2.0). Checkpoints: r = ref_r → -0.5 years; r = 0.5ref_r → +0.75 years; r = 0 → capped at +2.0 years.

[0063] VO2max (mutually exclusive index): The ACSM norm reference values ​​for age and gender stratification are shown in Table 2. Formula: AgeImpact_vo2 =clamp(−0.30 × (VO2_median − ref), −2.5, +2.5). Checkpoints: 10 units above the reference value → limit -2.5; equal to the reference value → 0; 10 units below the reference value → limit +2.5.

[0064] Resting heart rate (RHR): Baseline 62 bpm for men, 65 bpm for women; input the median RHR over 365 days. Formula: AgeImpact_rhr = clamp(0.15 × (RHR_median − ref_rhr), −1.0, +2.0). Checkpoints: Men 55 bpm → -1.0; 62 bpm → 0; 75 bpm → +1.95 years.

[0065] Nighttime HRV (RMSSD): Age-stratified norm reference values ​​are shown in Table 3. Formula: AgeImpact_hrv = clamp(−0.06 × (HRV_median − ref), −1.0, +1.5).

[0066] Total daily steps: <60 years old, reference 7000 steps / day; ≥60 years old, reference 5500 steps / day. Formula: AgeImpact_steps= clamp((ref_steps − steps_median) / ref_steps × 1.5, −0.5, +1.5).

[0067] High-intensity exercise duration (mutually exclusive metric): baseline 75 minutes / week, vig is the median weekly exercise over 28 days. Formula: AgeImpact_vig = clamp((75 − vig) / 75 × 1.5, −1.5, +1.5).

[0068] BMI segmented U-shaped formula: Optimal value 22 for <50 years old, optimal value 23 for ≥50 years old; dev = |BMI_median − optimal|. Standard interval 18.5 ≤ BMI < 25: AgeImpact_bmi = clamp(−0.3 + 0.12 × dev, −0.3, +0.5); Underweight BMI < 18.5: AgeImpact_bmi = min(+2.0, 0.4 × (18.5 − BMI)); Overweight 25 ≤ BMI < 30: AgeImpact_bmi = min(+2.0, 0.25 × (BMI − 25) + 0.3); Obese BMI ≥ 30: fixed +2.0 years.

[0069] Example 7: Complete Algorithm Verification

[0070] User basic information: Male, actual calendar age 40.9 years, stratified between 40-49 years old, all 10 indicators are valid, VO2max data is valid, high-intensity exercise duration is only used for demonstration and will be automatically removed during the aggregation phase. The baseline, reference standard, and complete calculation process for each indicator are shown in Table 5.

[0071] The total effect of effective indicators on age is approximately +2.39 years to +2.4 years; the original biological age is 40.9 + 2.4 = 43.3 years; last week's output was 42.9 years, Δ = +0.4, which did not exceed the speed limit ±0.5, and the final output is 43.3 years.

[0072] Aging rate calculation: Sleep debt is calculated using 7 / 28-day z-score, and other controllable indicators are calculated using 28 / 365-day z-score. The range-weighted improvement index I = -0.3 is obtained; Aging rate = clamp(1.0 + 0.3, -1.0, 3.0) = 1.3 ×, which means that the aging rate is 30% faster than the natural synchronous rate.

[0073] Attribution output example: Sleep debt +1.7 years → go to bed 30 minutes earlier daily; Sleep regularity +0.38 years → maintain a fixed schedule; Energy recovery +0.25 years → prioritize improving sleep; VO2max -0.9 years → maintain aerobic exercise. Sleep-related aging contributes 87% to the total aging, matching the sleep-dominated design principle.

[0074] Parameter Calibration Explanation: All indices, including slopes, upper and lower limits, and age-stratified population reference norms (P50 baseline values), are derived from publicly available data from authoritative international cohorts and clinical guidelines. Data sources include: the JAMA Population Physiological Aging Cohort Study, the Lancet Public Health Exercise and Sleep Epidemiology Review, the Psychophysiology Heart Rate Variability Standardization Trial, the ACSM American College of Sports Medicine Cardiopulmonary Function Classification Standards, and the WHO Global Physical Activity and Sleep Health Guidelines. All parameters are converted to age-biased slopes based on dose-risk response coefficients disclosed in the literature. Upper and lower limits for individual indices take into account clinical and physiological limits. After deployment, the zero-point calibration constant c_i can be fine-tuned based on the median P50 of the domestic population wearing samples. The core algorithm logic, such as the segmented mapping architecture, dual-time-window engine, collinear weight decay, and missing range redistribution, remains unchanged with parameter iterations.

[0075] The above description is merely a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. Within the scope of the technical concept disclosed in the present invention, equivalent substitutions for the technical solutions and features of the present invention are all within the scope of protection of the present invention.

Claims

1. A method of multi-dimensional biological age assessment by piecewise linear mapping, characterized in that, This method relies on multidimensional health data collected by wearable devices, combined with multi-period dynamic baselines, piecewise linear mapping, and a dual-time-window single-engine architecture, to output biological age and aging rate in parallel, including the following steps: Step S1: Data Collection and Preprocessing. Data on 10 health indicators across five dimensions—sleep, recovery, physical fitness, activity, and body composition—are collected. These 10 health indicators are: sleep debt duration, oversleep guard, sleep regularity score, energy recovery, VO2 max, resting heart rate, nocturnal heart rate variability, total daily steps, high-intensity exercise duration, and body mass index (BMI). A mutually exclusive calculation rule is set for VO2 max and high-intensity exercise duration, prioritizing VO2 max. If VO2 max is invalid, high-intensity exercise duration is automatically switched to. A combined IQR and 3σ outlier removal strategy is executed to clean the raw data. Step S2: Multi-period dynamic baseline calculation, setting three types of daily rolling time windows: 7 days, 28 days, and 365 days, calculating the median and standard deviation for each health indicator, and constructing the user's personal historical baseline; when the number of valid data days in the 365-day window is less than 90 days, the 28-day baseline is downgraded and marked as being in calibration. Step S3: Piecewise linear age impact mapping calculation. A piecewise linear mapping function is configured independently for each health indicator. The individual's historical baseline median is used as the input value, and the reference value of age x gender stratified population is used as the benchmark. Combined with the direction coefficient, slope and upper and lower limit rules, the indicator deviation is quantified into the age impact value. Step S4: Slow-layer output of biological age. Summarize the age impact values ​​of all effective indicators after weight adjustment, and add the user's actual calendar age to obtain the initial biological age. Apply the missing indicator range weight redistribution rule and the weekly change rate limit rule to output the biological age results on a weekly basis. Step S5: Output the fast aging speed layer, which is operated in parallel with Step S4 and uses the same piecewise linear mapping calculation engine. It calculates the improvement z-score for controllable leverage indicators, obtains the improvement index by weighting, and maps it to the aging speed multiplier. The output is updated daily. Step S6: Actionable Attribution Display. The 10 health indicators are divided into two categories: controllable leverage indicators and outcome indicators. Improvement suggestions for controllable leverage indicators are directly output, and the attribution results of outcome indicators are redirected to the corresponding upstream controllable leverage indicators.

2. The method for multi-dimensional biological age assessment by piecewise linear mapping according to claim 1, characterized in that, The joint outlier removal strategy described in step S1 is as follows: First, IQR removal, where data exceeding the range of [Q1-1.5xIQR, Q3+1.5xIQR] are identified as mild extrema and removed; Second, 3σ removal, where remaining data exceeding the range of [μ-3σ, μ+3σ] are identified as extreme extrema and removed; When the number of valid data points for an indicator does not reach the preset threshold, the indicator is identified as invalid, and missing data is left blank without interpolation.

3. The method for multi-dimensional biological age assessment by piecewise linear mapping according to claim 1, characterized in that, The baseline calculation described in step S2 includes two-level degradation rules: the 365-day window baseline is preferentially activated. If the number of days of valid data within the 365-day window is less than 90 days, the 365-day baseline is abandoned, and the 28-day time window baseline is uniformly used for calculation. The final result is marked as "calibrating" and the confidence interval is expanded; If the valid data in the 28-day window still do not meet the minimum valid number, it is marked as "seriously insufficient data" and the confidence interval is secondarily expanded.

4. The method for multi-dimensional biological age assessment by piecewise linear mapping according to claim 1, characterized in that, For the user's personal historical baseline in step S2, the learning rule for personal sleep demand is: personal sleep demand is learned through the median of 90-day rolling historical sleep duration, and the statistical caliber of sleep duration is the total duration from going to bed to falling asleep until getting up; new users in the transition period default to a standard sleep duration of 7.5h, the transition period is fixed at 90 days, and it will automatically switch to the personal learning baseline after the transition period expires.

5. The method for multi-dimensional biological age assessment using piecewise linear mapping according to claim 1, wherein, The general formula of the piecewise linear mapping function described in step S3 is: AgeImpact_i = clamp(slope_i × dir_i × (value_i − ref_i), floor_i, cap_i) + c_i Wherein, clamp(X, L, U) is a clamping function, which outputs L when X<L, outputs U when X>U, and outputs X when L≤X≤U; value_i is the personal long-term baseline median of the i-th indicator; ref_i is the age×gender stratified authoritative reference value corresponding to the i-th indicator; dir_i is a direction coefficient, taking a value of +1 or -1; slope_i is the slope, representing the age change amplitude corresponding to each unit of the indicator deviating from the reference value; floor_i is the lower limit of biological age reversal for the single indicator; cap_i is the upper limit of age increase for the single indicator; c_i is a neutral zero-point calibration constant.

6. The method for multi-dimensional biological age assessment using piecewise linear mapping according to claim 1, wherein, The reallocation rule for range weights of missing indicators described in step S4 is: let W_total be the sum of the ranges of all 10 indicators, W_valid be the sum of the ranges of current valid indicators, the adjusted age impact value of each valid indicator k is AgeImpact_k_adjusted = AgeImpact_k × (W_total / W_valid); when W_valid < W_total / 2, the system degrades to a partial estimation mode and expands the confidence interval; when the number of valid indicators N<4, the system rejects output and prompts insufficient data; the single-week change speed limit rule is: let the difference between the initially calculated biological age of this week and the output biological age of last week be Delta, Delta is clamped within the interval [-0.5, +0.5], the final output biological age = last week's biological age + clamp(Delta,-0.5, +0.5); For new users, only the baseline is calculated and no biological age is output in the first 4 weeks, and the first output is performed and the speed limit rule is activated after 4 weeks are completed.

7. The method for multi-dimensional biological age assessment by piecewise linear mapping according to claim 1, wherein, The calculation process for the aging rate mentioned in step S5 is as follows: calculate the improvement z-score for controllable leverage indicators, where the sleep debt duration is calculated using a 7-day baseline and a 28-day baseline, and the other controllable leverage indicators are calculated using a 28-day baseline and a 365-day baseline; calculate the weighted average with the range of each indicator as the weight to obtain the improvement index I; Aging rate = clamp(1.0 - kx I, -1.0, 3.0), where the sensitivity coefficient k is 1.0; when I>0, aging slows down, when I=0, it is in sync with natural aging, and when I<0, aging accelerates.

8. The method for multi-dimensional biological age assessment using piecewise linear mapping according to claim 1, wherein, Its features are, The mapping relationship between the outcome-type indicators and the upstream controllable leverage indicators in step S6 is as follows: Vitality recovery is redirected to sleep debt duration and sleep pattern score; maximum oxygen uptake is redirected to high-intensity exercise duration and total daily steps. Resting heart rate was redirected to duration of high-intensity exercise and sleep debt duration; nocturnal heart rate variability was redirected to sleep regularity score and duration of high-intensity exercise. When all controllable sleep-related indicators are missing, energy recovery and nocturnal heart rate variability are automatically redirected to controllable indicators related to exercise and body composition.

9. A multi-dimensional biological age assessment system of piecewise linear mapping, characterized by, To implement the method of claim 1, the system is characterized in that it adopts a dual-time-window single-engine architecture, comprising: The data acquisition and preprocessing module is used to collect raw data of multidimensional health indicators and perform two-level extreme value removal and data validity verification. The data acquisition and preprocessing module is used to collect raw data of 10 multidimensional health indicators, perform 3σ+IQR two-level extreme value removal and data validity verification, and output cleaned valid data. The multi-period dynamic baseline module calculates the median and standard deviation of each indicator based on daily rolling time windows of 7 days, 28 days, and 365 days, generates a personal historical baseline, and executes baseline downgrade judgment logic. The piecewise linear age impact calculation engine module has two independent read and write caches with built-in long window and short window. It calls the piecewise linear mapping function to calculate the age impact value for each indicator, providing a unified calculation logic for the two major output branches of biological age and aging rate. The biological age slow layer output module receives the output data from the calculation engine, combines the missing value weight redistribution rule and the weekly change rate limit rule, and generates and outputs the biological age results every week. The fast aging layer output module works in parallel with the slow biological age layer output module to calculate the improvement z-score, improvement index and aging speed multiplier, and generate and output the aging speed results daily. The module for fair handling and credibility labeling of missing values ​​completes the redistribution of range weights for missing indicators and calculates and labels the credibility of the results. The actionable attribution display module outputs either actionable actions or retargeting attribution suggestions based on the type of metric. The personalized benchmarking configuration module enables the learning of individual sleep needs, storage of individual recovery baselines, and automatic matching of age / gender stratified reference values.

10. The multi-dimensional biological age dynamic assessment system based on piecewise linear age effect mapping of claim 9, wherein, The piecewise linear age impact calculation engine module has two sets of independent read and write caches: a long window and a short window. The long window cache uses the median of the 365-day baseline and the median of the 28-day baseline for biological age calculation, while the short window cache uses the 28-day baseline and the 7-day baseline data for aging rate calculation. The two sets of cache data are isolated and do not overlap.

Images