Fractal Complexity of Daily Physical Activity Patterns Differs With Age Over the Life Span and Is Associated With Mortality in Older Adults

Abstract

Background

Accelerometers are included in a wide range of devices that monitor and track physical activity for health-related applications. However, the clinical utility of the information embedded in their rich time-series data has been greatly understudied and has yet to be fully realized. Here, we examine the potential for fractal complexity of actigraphy data to serve as a clinical biomarker for mortality risk.

Methods

We use detrended fluctuation analysis (DFA) to analyze actigraphy data from the National Health and Nutrition Examination Survey (NHANES; n = 11,694). The DFA method measures fractal complexity (signal self-affinity across time-scales) as correlations between the amplitude of signal fluctuations in time-series data across a range of time-scales. The slope, α, relating the fluctuation amplitudes to the time-scales over which they were measured describes the complexity of the signal.

Results

Fractal complexity of physical activity (α) decreased significantly with age (p = 1.29E−6) and was lower in women compared with men (p = 1.79E−4). Higher levels of moderate-to-vigorous physical activity in older adults and in women were associated with greater fractal complexity. In adults aged 50–79 years, lower fractal complexity of activity (α) was associated with greater mortality (hazard ratio = 0.64; 95% confidence interval = 0.49–0.82) after adjusting for age, exercise engagement, chronic diseases, and other covariates associated with mortality.

Conclusions

Wearable accelerometers can provide a noninvasive biomarker of physiological aging and mortality risk after adjusting for other factors strongly associated with mortality. Thus, this fractal analysis of accelerometer signals provides a novel clinical application for wearable accelerometers, advancing efforts for remote monitoring of physiological health by clinicians.

The growing use of wearable physical activity monitors holds great promise for clinical medicine (1). Through individual behavioral assessment, these devices may improve early disease detection and provide a powerful aid to precision medicine initiatives (2–4). However, because activity monitors are often used to simply track behavior, their utility as diagnostic tools remains unclear (2,3). While monitoring exercise participation may facilitate behavioral interventions, accelerometers in these devices also contain rich time-series data sets measuring overall patterns of movement that remain highly understudied. Here, we analyze wearable accelerometry data from a large, nationally representative U.S. sample of individuals ranging from 6 to over 85 years of age using a novel fractal method, and show that fluctuating patterns of daily activity can provide a physiologically relevant clinical endpoint that is associated with aging over the life span and is associated with mortality risk in community-dwelling older adults.

Our analysis takes advantage of the fact that the complexity of patterns in time-series data can provide a window into physiological function (5–7). Many human physiological processes, including heart rate and physical activity, fluctuate throughout the day in ways that can seem random (eg, fluctuating periods of movement and rest). However, recent work has shown that these fluctuations often possess an internal logic that resembles complex fractal patterns, and these patterns break down in disease states (5–8). In healthy physiological signals, peaks and valleys in intensity frequently vary such that fluctuations across long time-scales resemble those measured across shorter time periods, a characteristic known as self-affinity (5,9). In many disease states, self-affinity in physiological signals is reduced, and monitoring changes in complexity may provide an important method to identify early preclinical disease processes that predict survival (7,10–12). Interestingly, engagement in aerobic exercise reverses age-related loss of complexity in rodent motor activity (13), and in human heartbeat dynamics, gait, and overall patterns of physical activity (11,14–17), suggesting healthy lifestyle behaviors can beneficially impact physiological complexity.

Here, we use detrended fluctuation analysis (DFA) to perform a novel study of fractal complexity from wearable accelerometers that track physical activity during daily life. The DFA method is widely used to analyze nonstationary time-series signals across a variety of physiological systems and measures correlations between the amplitude of signal fluctuations across a range of time-scales (7,11,18–20). If fluctuations are strongly correlated across time-scales, the slope, α, relating the fluctuation amplitudes to the time-scales over which they were measured describes the complexity of the signal (Figure 1). Values of α close to 1.0 indicate a high degree of self-affinity and healthy fractal complexity (7). Values of α closer to 0.5 suggest randomness in the time-series fluctuations, and α values close to 1.5 indicate greater regularity in the signal (13).

Figure 1.

Example of detrended fluctuation analysis (DFA) in two subjects. (A and B) Accelerometer intensities for two subjects. (C) DFA analysis for subjects shown in A and B. Points show the change in fluctuation function across the range of time-scales used in the DFA analysis (from 10 minutes to 7 hours). Blue values show a subject with healthy fractal complexity (from A), indicated by a slope of the line relating F(t) to t of 1.0. Red values show a subject with reduced signal complexity (from B), indicated by a slope of 0.50.

While fractal analyses of activity may provide a promising avenue for the development of noninvasive biomarkers of disease, to date, human studies have only tracked small samples of individuals over a limited age range, and have not examined mortality as an endpoint. Using accelerometer data from over 11,600 participants in the National Health and Nutrition Examination Survey (NHANES), we sought to determine whether fractal complexity of physical activity patterns differs with age (acting as a biomarker of physiological senescence), whether these patterns are associated with exercise engagement, and whether fractal complexity is associated with mortality risk among older adults.

Materials and Methods

Actigraphy data were collected from the publicly available NHANES data set (9). NHANES is a large, stratified, multistage probability sample that is nationally representative of the community-dwelling U.S. population, spanning the life span from 6 to over 85 years of age (9). In NHANES, individuals older than 6 (mean age ± SD = 35.5 ± 23.5; see Supplementary Table 1) were asked to wear an ActiGraph AM-7164 accelerometer (ActiGraph, LLC, Fort Walton Beach, FL) on a belt around their waist for 7 days, except when bathing or sleeping (21). Accelerometer values are available for the 2003–2004 and 2005–2006 NHANES waves. We excluded subjects with less than 2 days of valid data (defined as data that were calibrated, returned with no reliability flags, and contained between 10 and 20 hours of valid wear time per day). Previous work has shown that 2 days of accelerometer data accurately capture physical activity levels and patterns in adults (22–24), and recent studies examining PA (25), and the association of accelerometer derived PA and mortality in the NHANES data set (26,27) have included participants with at least 1 day of valid wear time. We chose 2 days of wear time here as a conservative threshold for capturing PA patterns. We exclude days with greater than 20 hours of valid wear since these values indicate subjects likely wore accelerometers during sleep (28). Nonwear time was defined as any interval of 60 minutes or longer in which all accelerometer count values were 0, allowing for up to 2 minutes of count values between 0 and 100 (21). Periods of nonwear time were removed from the data set and data analysis below is performed across all valid days combined. Ma et al. (29) have shown that this method does not significantly alter the final results of the DFA analysis in time-series data sets. However, in addition to removing nonwear time from the data set, we performed DFA analyses for each day of each participant’s data set separately and calculated average parameters for all valid days. In addition, we performed a separate analysis where we included data from days for which individuals had 10 or more hours of consecutive wear time and averaged parameters from each day. Results for these analyses are similar to the analysis presented below and are available in Supplementary Materials. The final data sets used in the DFAs were device intensity values for each 1-minute epoch of weartime (n = 11,694; n_male = 5,731; n_female = 5,963).

DFAs were performed on each subject using the nonlinearTseries package in R version 3.3.1 (30). This method is described in detail elsewhere (7,20) (see Figure 1). Briefly, in a DFA, time-series data are first integrated, following the removal of the global mean of the signal. Next, the time-series is divided into a series of nonoverlapping windows of some size, t. Finally, the root mean square of residuals from a least-squares regression of the time-series within each window (detrending) is calculated to compute the fluctuation amplitude, F(t). These steps are then repeated across a series of 25 window sizes ranging from 10 minutes to 7 hours in length. The fractal complexity of the time-series is determined by the slope, α, of the least-squares regression line relating F(t) to window size (t). If α = 0.5, the time-series does not include long-range correlations in fluctuations, suggesting fluctuations are mainly white noise. If 0.5 <α ≤ 1.0, the time-series contains positive long-range correlations in fluctuations, indicating increasing fractal complexity as α nears 1.0. If α > 1.0, the time-series begins to have more regular or predictable fluctuations as α approaches 1.5. Thus, in analyses of motor activity data, α values closer to 1.0 are considered representative of healthy fractal physiological complexity (20).

We analyzed changes in α between sexes and across age groups. In addition, we examined the association of physical activity with α. Physical activity was calculated as time spent in moderate-to-vigorous physical activity (MVPA) from the actigraphy records, using the package nhanesaccel (31) in R version 3.3.1 (30). We used the following cut-points to define MVPA intensity levels: MVPA₁: ≥760 counts (32), and MVPA₂: ≥ 2,020 counts (21). The lower cut-point for MVPA₁ may reflect a better estimate of moderate-to-vigorous intensity levels for older adults (32). In addition to analyses of MVPA, we also examined other measures of PA in our survival analyses: sedentary: <100 counts, light intensity: between 100 and 759 counts, and total counts per day (results of these analyses are presented in Supplementary Materials). For these analyses, we first calculated time spent in a particular activity level as a percentage of total wear time. Next, we calculated age-specific quartiles of percent of time in each activity level using 10-year age increments.

Mortality data were taken from the linked death records in the National Death Index through December 31, 2011 (33). For analyses relating α and mortality, we controlled for the presence of the following chronic diseases: congestive heart failure, coronary heart disease, stroke, diabetes, cancer, mobility limitations (defined as difficulty walking ten steps or difficulty walking a quarter mile), angina, emphysema, chronic bronchitis, liver disease, kidney disease, self-reported confusion or memory problems, or arthritis. We restricted our analysis to individuals aged 50–79 years because previous work has shown that the functional form of the hazard of death differs in those aged 79 and younger compared with those aged 80 years and older (26). Based on these criteria, the sample for mortality analyses included 2,712 individuals and 329 deaths (see Supplementary Materials for analyses including subjects >80 years of age).

Statistical Analyses

Generalized linear models were used to examine the effects of covariates on the DFA exponent. Generalized linear models were adjusted for accelerometer wear time, and took into account sampling weights and the complex survey design of NHANES using the survey package in R, version 3.3.1 (30). For survival analyses, Cox proportional hazard models were used to evaluate the effects of DFA variables on mortality by estimating the hazard ratio (HR) and associated 95% confidence interval (CI) in subjects aged 50–79 years and in subjects aged 50 years and older at the time of the examination. To evaluate the effects of DFA variables after controlling for potential confounders, five adjusted models were fitted: (i) adjusted for age, (ii) adjusted for age and PA intensity quartile, (iii) adjusted for age, PA intensity quartile, and presence of chronic conditions, (iv) adjusted for age, PA intensity quartile, presence of chronic conditions, and accelerometer wear time, and (v) adjusted for age, PA intensity quartile, presence of chronic conditions, accelerometer wear time, and other factors that were associated with mortality in univariate models (see Table 2 and Supplementary Table S5). The proportional hazard assumption was assessed using both graphical and time-dependent variable approaches. Survival analyses were performed using Proc SurveyPhreg and Proc Phreg in SAS 9.4. Four-year adjusted mobile examination center weights were used in statistical analyses because two waves of NHANES were pooled, and strata and primary sampling units were used to account for complex survey characteristics, as recommended by NHANES (34). All tests were two-sided, with the significance level set at 0.05. The comparisons of model fit were based on the approach described in Lumley (35). Specifically, sampling weights were first rescaled to sum equal to the sample size. All models were then tested with the rescaled weights and with additional covariates representing strata and primary sampling units to obtain Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and likelihoods for model comparisons. Finally, c-statistics based on the unweighted data were calculated to determine the discriminatory ability of a given model. Following Studenski et al. (36), an increase of 0.025 in the c-statistic was considered a clinically meaningful enhancement in discriminatory ability.

Table 2.

All-Cause Mortality Hazard Ratios for DFA Scaling Exponent

MVPA Cut-point	Model	Adj HR	95% CI	p Value	AIC	c-statistic
MVPA1	1a	0.6	0.48–0.75	<.0001	3,556.7	0.721
	2b	0.73	0.60–0.89	<.01	3,520.0	0.739
	3c	0.75	0.62–0.92	<.01	3,494.7	0.749
	4d	0.76	0.62–0.92	<.01	3,495.4	0.749
	5e,f	0.75	0.59–0.97	.03	684.2	0.77
MVPA2	1a	0.6	0.48–0.75	<.0001	3,556.7	0.721
	2b	0.66	0.53–0.83	<.001	3,545.0	0.733
	3c	0.67	0.54–0.84	<.01	3,517.3	0.741
	4d	0.68	0.54–0.85	<.01	3,517.9	0.742
	5e,f	0.64	0.49–0.82	<.01	680.5	0.784

Notes: MVPA quartiles are calculated within each age category. Two accelerometer thresholds were used to account for time spent in MVPA: 760 CPM for MVPA₁, and 2020 CPM for MVPA₂.

AIC = Akaike Information Criterion; CI = Confidence interval; DFA = Detrended fluctuation analysis; MVPA = Moderate-to-vigorous physical activity.

^aControlling for age. ^bControlling for age and MVPA quartile. ^cControlling for age, MVPA quartile, and chronic conditions. ^dControlling for age, MVPA quartile, chronic conditions, and weartime. ^eControlling for age, MVPA quartile, chronic conditions, weartime, sex, marital status, education, difficulty walking, hypertension, and total cholesterol. ^f2,069 individuals with missing information on at least one of the variables.

Results

The DFA exponent varied significantly with both age and sex (Figure 2, Table 1). Specifically, in both men and women, α became significantly shallower in older individuals, signaling a reduction in fractal complexity, and thus providing a strong biomarker of aging. Additionally, there was a significant interaction between age and sex, whereby women had a greater age-related decline in complexity than men (Table 1). PA patterns that are associated with age and sex are not simply a function of overall activity. The DFA exponent (α) was weakly positively correlated with overall activity, measured as average total daily counts (p < .001, r² = .05) and was weakly correlated with the number of days participants wore the accelerometers (p < .001, r² = .02). However, given the small amount of variation in α explained by these variables, α appears to be capturing a pattern of activity that differs from overall activity.

Figure 2.

Changes in detrended fluctuation analysis (DFA) exponents with age and sex. Scaling exponent α from DFA shown in 10-year age groups for men (orange) and women (teal). Scaling exponents (unadjusted for covariates in Table 1) change with both age and sex, and there is a significant interaction between age and sex with women showing lower exponents at all ages and a greater decrease in α with age (see Table 1).

Table 1.

Effects of Age, Sex, and MVPA on DFA Scaling Exponents

MVPA Variable	Factors	Estimate	SE	t Value	p Value
MVPA1	(Intercept)	8.38E−01	1.61E−02	52.072	<2E−16
	Age	−1.76E−03	2.05E−04	−8.601	1.21E−08
	Sex	−4.47E−02	6.44E−03	−6.936	4.53E−07
	MVPA Quartile	−1.13E−02	4.79E−03	−2.364	.0269
	Age × Sex	3.87E−04	8.88E−05	4.357	.0002
	Age × MVPA Quartile	4.21E−04	6.40E−05	6.585	1.02E−06
	Sex × MVPA Quartile	7.68E−03	2.40E−03	3.196	.0040
MVPA2	(Intercept)	7.39E−01	1.78E−02	41.541	<2.00E−16
	Age	−1.53E−03	2.36E−04	−6.483	1.29E−06
	Sex	−3.77E−02	8.45E−03	−4.46	1.79E−04
	MVPA Quartile	1.00E−02	6.36E−03	1.575	.13
	Age × Sex	2.75E−04	9.83E−05	2.793	0.01
	Age × MVPA Quartile	3.52E−04	6.53E−05	5.388	1.79E−05
	Sex × MVPA Quartile	1.53E−02	2.97E−03	5.167	3.09E−05

Notes: MVPA quartiles are calculated within each age category. Model was adjusted for accelerometer wear time and takes into account complex survey design. Two accelerometer thresholds were used to account for time spent in MVPA: 760 CPM for MVPA₁, and 2020 CPM for MVPA₂.

DFA = Detrended fluctuation analysis; MVPA = Moderate-to-vigorous physical activity.

The effects of time spent in MVPA on α depend on the accelerometer threshold used. For the lower MVPA threshold (MVPA₁), time spent in MVPA was significantly associated with α, while MVPA₂ alone did not explain significant variation in α. For both measures of MVPA, variation in α was associated with significant interactions of age-specific MVPA quartiles with age and with sex (Table 1). Within each age category, individuals in the higher age-specific MVPA quartiles generally had higher DFA exponents (α closer to 1.0) (Figure 3, Table 1). Although there are differences across age and MVPA categories, the putative beneficial effects of higher levels of activity on DFA were most consistently observed after age 50 (Supplementary Table S2), and in women compared with men (Supplementary Table S3).

Figure 3.

Changes in detrended fluctuation analysis (DFA) exponents with participation in moderate-to-vigorous physical activity (MVPA). Within each 10-year age category, individuals engaged in higher levels of MVPA (based on age-specific quartiles where time spent in MVPA as a percentage of wear time increases with increasing quartiles) have higher scaling exponents α (values shown here are not adjusted for covariates found in Table 1). MVPA is calculated here using the accelerometer threshold of 2020 CPM.

Fractal complexity is strongly associated with mortality in this sample over a range of follow-up from 6 to 9 years (33). In adults aged 50–79 years, a lower α value is associated with increased mortality (Table 2). This result holds true when adjusting for accelerometer wear time, time spent in MVPA using either threshold, and other variables that influence mortality, including education level, marital status, chronic diseases, hypertension, and total cholesterol levels. Despite established differences in the functional form of the hazard of mortality in those over 80 years of age (26), results also remain significant when including participants over the age of 80 (see Supplementary Table S5). The association between the DFA exponent and mortality remains strong when we include alternative variables to characterize activity levels (eg, time spent sedentary, in light activity, or total counts per day; see Supplementary Tables S6–S8), and results are similar if we exclude individuals who died in the first year of follow-up (Supplementary Table S9). Finally, the association between the DFA exponent and mortality in this sample is robust to different methods of processing accelerometer data during the DFA analysis, including calculating α as the average DFA exponent for each valid day (Supplementary Tables S10 and S11). Together, these findings suggest that the association between α and mortality is robust and is not explained by chronic illness or by the oldest individuals in the sample. Adding the DFA exponent to a survival model including only age increases the c-statistic by a clinically meaningful amount (unadjusted hazard ratio [HR] for age = 1.09 [95% CI: 1.07–1.11], c-statistic_age = 0.684 vs c-statistic_{age + α} = 0.721). In the fully-adjusted model, every 0.1 unit increase in α is associated with a 36% reduction in mortality risk. Thus, there is a 180% difference in mortality risk in individuals with activity patterns exemplified in Figure 1 who have values of α associated with random activity fluctuations (α = 0.5) and α values associated with high signal self-affinity (α = 1.0).

Discussion

This is the largest study, to date, examining fractal measures of accelerometry data from wearable monitors, and the first to demonstrate the association between fractal complexity of physical activity patterns and mortality. Based on these results, we suggest that the DFA exponent α represents a novel biomarker of all-cause mortality risk and may be of great value in clinical settings and applied research as a predictor of subsequent health status. Furthermore, the negative association of α with age in the full sample provides a potentially new way to track physiological senescence. Thus, our results may advance the use of noninvasive and cost-effective remote measures of physiological complexity that can identify those at the greatest risk for illness and may help evaluate interventions.

Numerous studies have shown that engagement in physical activity is associated with lower mortality risk in the NHANES sample (27,36–40). In general, PA-related risk reductions are attributed to the effects of exercise on cardiometabolic disease biomarkers, inflammation, immune function, and weight reduction (40). Our results suggest an alternate pathway for reducing mortality risk through engagement in PA where participation in MVPA may alter fractal activity patterns associated with mortality. In contrast to the general age-related reductions in α, we have shown that engagement in higher levels of MVPA is associated with healthier patterns of complexity in motor activity in older adults and in women across much of their adult life span. Although it is too early to know whether exercise interventions can improve DFA exponents, recent work has shown that aerobic exercise can enhance fractal complexity in both heartbeat and gait dynamics in humans (14–17). In older adults, higher levels of physical activity (steps per day) are associated with higher DFA exponents measured from wearable accelerometers worn for 2 weeks (11). Additionally, experimental work suggests that exercise can reverse the effects of aging on fractal complexity of motor activity in rodents (13). In fact, exercise restriction in rodents leads to a loss of fractal complexity in both young and old individuals (13), similar to our human data, and data from Cavanaugh and colleagues (12), showing patterns of reduced complexity in individuals engaged in the lowest quartiles of exercise across age groups. The sex differences in DFA exponents in our study are noteworthy, and it is possible that these differences are due, in part, to overall lower levels of MVPA in female participants. In our sample, DFA exponents in women appear to be especially sensitive to engagement in higher levels of MVPA, and thus the lower overall levels of MVPA in women may explain, at least in part, their lower DFA exponents across the sample (Supplementary Table S3). Others have shown that women across all age groups engage in lower levels of MVPA than men in the NHANES sample (21), suggesting that exercise interventions tailored to enhance fractal complexity in women during aging may be an important direction for future studies. In our sample, we also found a relationship between sex and age-specific MVPA quartiles (females < males; p < 2.0E−16; Supplementary Table S4) and we found a significant interaction between sex and MVPA on the DFA exponent (Table 1). While randomized controlled trials are needed to confirm the effects of MVPA on fractal physiological complexity, our results suggest a key benefit of engaging in regular MVPA and a novel way to identify specific subgroup populations who may preferentially gain from targeted exercise interventions. For example, for older adults and women over the adult life span, interventions that shift individuals into higher age-specific quartiles of MVPA may provide the greatest benefits for fractal complexity, and thus for overall health.

Although the underlying biology of motor activity complexity remains unclear, mounting evidence links fractal complexity in activity patterns to circadian rhythm (CR). Previous work suggests that the complexity of motor activity patterns is controlled to some degree by the suprachiasmatic nucleus (SCN) (41). In rodent models, lesioning the SCN leads to a break down in the pattern of daily activity fluctuations measured by DFA (41), and in humans, benign tumors that compress the SCN lead to reductions in α compared to control participants (32). Since the SCN is responsible for maintenance of CR across physiological systems (42), it is possible that the fractal complexity of motor activity is associated with CR. In fact, the connection between fractal complexity of activity and CR may help explain why participation in MVPA is associated with higher DFA scaling exponents in women and older adults. Voluntary exercise is known to enhance CR in humans and nonhuman animal models, and it is possible that exercise affects the SCN in ways that increase the DFA scaling exponent (13). Interestingly, Alzheimer’s disease (AD) has been associated with reduced fractal complexity of physical activity (ie, reductions in α) over and above age-related decreases, and AD is also known to negatively affect CR (19,20,43). Hu et al (19,43). showed that reduced DFA scaling exponents in accelerometry data are found in individuals with dementia, scaling exponents are correlated with cognitive performance, and that the breakdown in self-affinity of activity patterns is exacerbated in individuals with severe dementia effects on the SCN and high levels of amyloid plaque deposition in the brain. However, traditional measures of CR are strongly impacted by external environmental cues, such as variance in the scheduling of daily activities (43). Fractal analyses of motor activity are not affected by external cues and seem to better reflect underlying physiology (43). Thus, changes in fractal patterns may be a key indicator of SCN functioning, and CR regulation, across populations.

Our study has several strengths that support the use of DFA exponents as a novel biomarker of morbidity and mortality. This is the largest study, to date, examining fractal measures of accelerometry data from wearable monitors, and our results are based on a nationally representative community-dwelling sample assessing activity complexity from childhood to old age. By analyzing both an age-restricted sample (50–79) and the full sample of older adults (>50), we show that the value of the DFA exponent is not driven by effects in the oldest old. Additionally, the prospective cohort design and the relatively long follow-up period for mortality data (6–9 years) is an important strength of this sample. While promising, our study does have limitations. Although we controlled for the major chronic illnesses and conditions of aging, as an observational study that is subject to potentially unmeasured confounds, our findings indicate the need for future follow-up studies using an experimental approach to both determine the predictive power of the DFA exponent for disease-specific morbidity, and to better understand the role of MVPA and/or other lifestyle factors in improving fractal complexity of daily activity. Additionally, NHANES participants wore accelerometers for up to 7 days, however future studies have the potential to track fractal complexity of physical activity over much longer time periods, likely enhancing the utility of this measure, and helping us understand how activity patterns at younger ages affect physiological aging and senescence (44).

Despite these limitations, we believe there is ample evidence that the DFA exponent is a powerful biomarker of physiological health that is associated with mortality risk in adults. Given its noninvasive nature, and the growing ubiquity of accelerometers in society (eg, smart phones and watches), we suggest that the use of wearable technology combined with novel analytic approaches has great potential in tracking both physiological and behavioral aspects of everyday activity, with the opportunity to enhance diagnostics and predictors of health over the life span. The application of such wearable devices in clinical settings may provide a way to identify developing disease at the earliest stages, when treatments and preventative therapies may be most effective. In this way, low-cost, noninvasive wearable technologies may provide a means of monitoring the complexity of activity across the life span, supporting current efforts to advance health through individualized precision medicine and prevention initiatives.

Funding

This work was supported by the National Institute on Aging (AG019610); the National Science Foundation (BCS 1440867); the State of Arizona and Arizona Department of Health Services (ADHS); Ken and Linda Robin; and the McKnight Brain Research Foundation.

Conflict of Interest

None reported.