Mobility Abilities Mediate the Association of a More Active Lifestyle With Mobility Disability in Older Adults

Abstract

Background

Few studies have analyzed sensor-derived metrics of mobility abilities and total daily physical activity (TDPA). We tested whether sensor-derived mobility metrics and TDPA indices are independently associated with mobility disabilities.

Methods

This cohort study derived mobility abilities from a belt-worn sensor that recorded annual supervised gait testing. TDPA indices were obtained from a wrist-worn activity monitor. Mobility disability was determined by self-report and inability to perform an 8-feet walk task. Baseline associations of mobility metrics and TDPA (separately and together) were examined with logistic regressions and incident associations (average 7 years follow-up) with Cox models. Mediation analysis quantified the extent mobility metrics mediate the association of TDPA with mobility disability.

Results

724 ambulatory older adults (mean age 82 years, 77.4% female) were studied. In separate models, mobility abilities (eg, step time variability, turning angular velocity) and TDPA were related to mobility disabilities. Examined together in a single model, mobility abilities remained associated with mobility disabilities, while TDPA was attenuated. This attenuation of TDPA could be explained by mediation analysis that showed about 50% of TDPA associations with mobility disabilities is mediated via mobility abilities (prevalent mobility disability 54%, incident mobility disability 40%, incident loss of ambulation 50%; all p’s < .001).

Conclusions

Sensor-derived mobility metrics assess more diverse facets of mobility. These metrics mediate approximately half of the association of higher levels of daily physical activity with reduced mobility disability in older adults. Findings may inform the design of targeted interventions to reduce mobility disability in late life.

Mobility disability and loss of ambulation are key milestones that lead to the loss of independent living in older adults. Conventional gait speed is commonly used to assess gait in older adults and is a robust, but nonspecific, predictor of mobility disability and other adverse health outcomes (1). Slowed gait can be caused by deficits of diverse mobility abilities needed to ensure normal gait and balance during everyday living (2,3). Studies by many investigators, as well as our own prior work, have demonstrated that the widespread availability of low-cost unobtrusive sensors offers the potential for quantifying gait speed as well as a wide array of mobility abilities that can lead to mobility disability (1,4–7). Measuring quantitative mobility metrics may also lead to earlier detection of adults at risk for mobility disability and facilitate targeted therapies that maintain ambulation and independent living in aging adults.

Unimpaired mobility abilities may not be sufficient to prevent the development of mobility disability. There is increasing recognition of the importance of an active lifestyle, especially higher total daily physical activity (TDPA), for maintaining mobility and slowing the development of mobility disability (8,9). Mobility abilities reflect the capacity to move, while TDPA represents the quantity of actual movement during everyday living. Mobility abilities and TDPA are related, but distinct phenotypes, as mobility abilities only account for approximately 16% of the variance in TDPA (10). Since TDPA is dependent on the volitional decision to move, adults with unimpaired mobility abilities may choose not to move, and conversely, adults with impaired mobility abilities may, nonetheless, be quite active and show high levels of TDPA. Just as sensors can be employed to record mobility abilities that are not captured using traditional motor performances, low-cost activity monitors provide the means to quantify TDPA, circumventing the limitations of self-report physical activity questionnaires or recall bias that may affect older adults (11). Further, mobility disability can be defined in a variety of ways, for example, using self-report instruments or objective abilities and performances. Elucidating which mobility abilities relate to different mobility disability phenotypes may aid risk stratification and the development of targeted therapies for specific abilities.

The current study builds on our prior studies in the Rush Memory and Aging Project cohort that have deployed different sensors in the community setting and have shown that sensor-derived metrics of mobility abilities and TDPA are both associated with incident self-report mobility disability (5,6,12). Our prior studies did not examine the joint contributions of quantitative sensor-derived mobility metrics and TDPA with mobility disability, however. The current study seeks to test the hypothesis that since sensor-derived mobility abilities and TDPA indices are distinct weakly related phenotypes that they would be independently associated with mobility disability.

Method

Participants were from the Rush Memory and Aging Project (MAP), an ongoing longitudinal, clinical-pathologic cohort study of aging that began in 1997 (13,14). Briefly, participants were older adults recruited from retirement communities and subsidized senior housing facilities across Chicago and Northeastern Illinois. Participants are free of diagnosed dementia upon enrollment, and they agreed to annual study evaluations and to organ (brain, spinal cord, muscle, and nerve) donation. All study procedures were approved by a Rush University Medical Center Institutional Review Board and administered in accordance with the Declaration of Helsinki. All participants provided written informed consent and an Anatomic Gift Act at enrollment.

Study recruitment and enrollment occur on a rolling basis. Participants complete annual study evaluations in their homes which assess cognition, motor function, comorbidities, experiential and personality risk factors, and disability (including mobility disability and ability to ambulate). Wrist-worn activity monitors to quantify TDPA were added in 2005. Belt-worn sensor recordings of annual gait and balance testing were added in 2012.

The analytic baseline for the current study was based on the first cycle that a MAP participant had measures of both sensor mobility abilities and TDPA. In addition, eligible participants were ambulatory at the analytic baseline (ie, able to walk 8 feet, a requirement for mobility testing) and had also had a valid self-reported mobility disability score. In total, 1 141 participants had data on mobility abilities while 1 392 had data on total daily activity, though 728 had measures of both mobility abilities and total daily activity in the same testing cycle. Of these 728 participants, 1 was missing data on self-reported mobility disability, and 3 were unable to walk 8 feet, resulting in an overall analytic sample size of 724 participants. Of these, 33 participants did not have longitudinal data (eg, one additional assessment) for self-reported mobility disability or ambulatory status and were only included in cross-sectional analyses (N = 20 died before their second visit and N = 13 lacked sufficient follow-up data). For longitudinal analyses, the average follow-up was approximately 7 years.

Assessment of Mobility Abilities

Device

During testing of gait and balance, participants wore a sensor (Dynaport MT, McRoberts B.V., The Hague, the Netherlands) positioned on their lower back and secured with a neoprene belt. This device contains 3 orthogonally oriented piezoelectric accelerometers and 3 gyroscopic sensors allowing for 100 Hz sampling of 3-dimensional acceleration and rotation rate of the lower trunk.

Gait testing procedures

Gait and balance testing obtained at participants’ home included 3 performance-based mobility tasks: a 32-foot walk, standing balance, and a timed-up-and-go (TUG) test. Because space is generally limited in participant homes compared with clinic settings, the 32-foot walk was completed on an 8-foot course where participants walked at their self-selected pace back and forth twice without stopping, rather than only an 8-foot course which would have captured fewer steps. Only metrics from the straight walking portion are extracted from the 32-foot walk data (described below) in this study. For the standing balance test, participants were asked to stand in a comfortable position for 20 seconds with their eyes closed. For the modified TUG, participants were instructed to stand up from a chair, walk 8 feet at a comfortable self-selected pace, turn around, return to the chair, and turn to sit down.

Data collection of mobility recordings through a laptop computer was controlled by a research assistant via a Bluetooth hand-held clicker. All 3 performances were recorded in a single file, with markers automatically embedded in the file to identify the beginning and the end of each performance. Mobility recordings were uploaded to the RADC central server and analyzed later using custom-developed software as previously published (15).

Extraction of sensor-derived mobility abilities

We adapted previously published validated formulas for automated analysis of the large numbers of mobility recordings that were collected (15). These analytic tools were incorporated into a pipeline integrating data collection and the extraction of sensor-derived mobility metrics.

Sixteen different summary scores were derived based on principal component analysis and review of the past literature to quantify each of the different performances or subtasks listed in Table 1 and detailed in Supplementary Materials (7,10,15). The 32-foot walk (straight walking portions only) was summarized by 4 scores including: gait regularity, cadence, pace, and variability of mean step time. Standing balance was summarized by 3 scores: sway frequency, sway jerk, and sway magnitude. The TUG is a complex task that requires the successful completion of 6 individual movements or subtasks including standing, first walk, first turn, second walk, second turn, and sitting. The TUG walk components were quantified by 4 scores: regularity, cadence, walking pace, and variability of the mean step time. The TUG turns were summarized by yaw magnitude (ie, trunk angular velocity). Transition from sitting to standing was summarized by 2 scores: complexity of stand-up and duration from the TUG stand-up. Transition from standing to sitting was quantified by 2 scores: descension time and smoothness.

Table 1.

Mobility Tasks, Abilities, Mean Baseline Z-Scores, And Correlations With TDPA

Mobility Tasks	Mobility Abilities	Mean (SD)	Spearman Correlation With Total Daily Physical Activity*
32-Foot walk	Regularity	0.10 (0.84)	0.34
	Cadence	0.15 (0.90)	0.21
	Pace	0.17 (0.96)	0.40
	Step time variability	−0.09 (0.92)	−0.16
Standing balance	Sway frequency	0.55 (1.07)	NS
	Sway jerk	0.79 (1.13)	−0.13
	Sway magnitude	0.51 (0.98)	−0.11
TUG walk	Regularity	0.12 (0.86)	0.23
	Cadence	0.10 (0.98)	0.25
	Pace	0.23 (1.03)	0.42
	Step time variability	−0.09 (1.0)	−0.17
TUG turn	Yaw magnitude	0.01 (0.91)	0.49
TUG stand-up	Complexity	−0.12 (0.67)	0.25
	Duration	0.05 (0.84)	−0.33
TUG sit-down	Descension	−0.10 (0.88)	0.26
	Smooth	−0.29 (0.79)	NS

Notes: TDPA = Total Daily Physical Activity; TUG = timed up and go.

*All Spearman rank-order coefficients are statistically significant at p < .05, except when indicated by NS.

Assessment of TDPA

Participants wore an activity monitor (Actical; Philips Healthcare, Bend, OR) on their nondominant wrist continuously for up to 10 consecutive days. The device was set to record activity every 15 seconds. The device averaged all activities for each 15-second duration (epoch) and saved the average physical activity counts for each epoch. The activity counts of recording epochs of a complete 24-hour recordings were summed to calculate total daily activity counts, which were averaged across full days of recordings to yield TDPA (16,17).

Assessment of Mobility Disability

The definition of mobility disability depends on the mobility performance or instrument used to quantify mobility. We examined self-reported mobility disability and a loss of ambulation based on clinical assessment of the inability to ambulate an 8-foot walk that we have employed in prior studies.

Self-reported mobility disability

Self-reported mobility disability was assessed using the Rosow-Breslau measure, which considers the ability to (1) do heavy housework, (2) walk up and down stairs, and (3) walk ½ mile (18). For each task, participants are asked whether they can complete the task independently, need assistance, or if they are unable to do it. Participants who report needing assistance or are unable to perform any activity are categorized as having mobility disability. Self-reported mobility disability was characterized at the analytic baseline visit and all subsequent visits thereafter to determine prevalent and incident self-reported mobility disability, respectively.

Assessment of loss of ambulation

At each visit, participants were examined for the ability to do an 8-foot walk. All participants were able to walk at least 8 feet at the analytic baseline. Incident loss of ambulation was defined as the first visit where the participant was unable to complete an 8-foot walk (19).

Other Covariates

We examined a variety of other factors that can be related to motor function and disability. Covariates obtained via self-report were recorded at study entry and included date of birth, sex, race, and years of education. Near vision acuity was measured using the Rosenbaum Pocket Vision Screener. We characterized participants as having vision impairment if their best-corrected binocular visual acuity was worse than 20/40 (20). Body mass index (BMI) was calculated using measured height and weight. Depressive symptoms were assessed using a 10-item version of the Center for Epidemiologic Studies Depression scale (21). Vascular risk factors (sum of hypertension, diabetes, and current/former smoking) and number of vascular conditions (sum of claudication, heart conditions, congestive heart failure and stroke [also determined from neurologic evaluation]) were collected via self-report.

Statistical Analyses

Descriptive statistics were calculated for the sample, including the mean and standard deviation for baseline age, total daily activity, and z-scores for each of the mobility abilities, as well as frequencies for sex and race. We also calculated Spearman rank-order correlation coefficients for the correlations between total daily activity and each of the mobility abilities.

The goal of the current analyses was to test whether mobility abilities and TDPA are independently associated with mobility disabilities. In prior publications, we reported that sensor-derived mobility metrics and TDPA examined in separate models were associated with mobility disability (6,7). To confirm these prior reports in our current analytic cohort, we replicated the prior findings, which are described in Supplementary Material (see Supplementary Statistical Analysis and Supplementary Tables S2–S4).

To test our primary hypothesis, our primary model included both sensor-derived mobility abilities and TDPA together in a single model to determine if both were independently related with each of the 3 disability outcomes. These models were adjusted for age, sex, race, and education, and final models including both mobility abilities with TDPA were also further adjusted for visual impairment, BMI, depressive symptoms, vascular risk factors, and vascular diseases.

Our primary analysis showed that mobility abilities remained associated with mobility disabilities when TDPA was included in a single model. Yet, contrary to our initial hypothesis, the associations of TDPA with mobility disability measures were attenuated when mobility abilities were included in the same model. These results led us to employ mediation analysis to test if mobility abilities mediated the association of TDPA with mobility disability outcomes. The direction of our hypothesized associations for the variables included in the mediation model is shown in Figure 1, with the mediation analysis quantifying the proportion of the direct association between TDPA and mobility disability, and the proportion of the indirect association mediated through the sensor-derived mobility abilities. First, we examined the mediation effect of each of these mobility abilities with TDPA in separate mediation analysis. Then, we examined the mediation effects of multiple mobility abilities together in a combined model for each of the disability outcomes, using the regression methods described by VanderWeele and Vansteelandt (22). All mediation analyses were conducted using R package Mediation (23) and were fully adjusted for age, sex, race, years of education, visual impairment, BMI, depressive symptoms, vascular diseases, and vascular risk factors. Descriptive statistics and all other models were run using SAS software version 15.2 (SAS Institute, Cary, NC). Statistical significance was determined at α of 0.05 except where otherwise noted.

Figure 1.

Mediation model depicting the direct effect of total daily activity on mobility disability and the indirect effect through mobility abilities.

Results

Baseline Characteristics of the Analytic Cohort

There were 724 older adults included in these analyses; the mean age was 81.6 (standard deviation [SD]: 7.5) years, 76.1% were female (N = 551), 91.9% were white race (N = 665), and the mean years of education was 15.5 (SD: 3.0). Mean total daily activity in counts per day was 2.05 × 10⁵ (SD: 1.19 × 10⁵). Means for each of the mobility abilities as well as correlations between mobility abilities and TDPA are listed in Table 1, showing zero to modest correlations. Correlations between mobility abilities and TDPA were weak to moderate and are shown in Supplementary Table S1.

Mobility Abilities and Mobility Disability Outcomes

Prevalent self-reported mobility disability

Among the full sample of 724 participants, 53% (N = 384) had mobility disability at baseline. In separate models, all mobility abilities from the 32-foot walk, sway magnitude from standing balance, and all mobility abilities derived from the TUG aside from the smooth measure from the TUG sit-down were separately associated with odds of self-reported mobility disability, adjusting for age, sex, and race (p < .05 for each; Supplementary Table S2). When considered together, 6 mobility abilities—all from the TUG—were independently associated with odds of self-reported mobility disability.

Incident self-reported mobility disability

There were 332 of 724 (45.9%) participants who were free of self-reported mobility disability at baseline and had at least one follow-up visit, among which 68% (n = 225) developed incident self-reported mobility disability over an average of 6.9 (SD: 2.6) years of follow-up. With few exceptions, most mobility abilities associated with prevalent self-reported mobility disability were also associated with time-to-incident self-reported mobility disability (Supplementary Table S3). In the final combined model, only sway magnitude from the standing balance measure as well as TUG stand-up time, step time variability, and yaw magnitude were independently associated with incident self-reported mobility disability.

Incident loss of ambulation

Of the 688 participants with longitudinal data on ambulation, (by definition, all participants were ambulatory at analytic baseline), 14% (N = 97) developed loss of ambulation over an average of 7.3 (SD: 2.7) years of follow-up. Similarly, most mobility abilities associated with prevalent self-reported mobility disability were also associated with time-to-incident loss of ambulation (Supplementary Table S4). In final, combined models, standing balance sway jerk as well as TUG walking pace and yaw magnitude were independently associated with incident loss of ambulation.

Total Daily Physical Activity and Mobility Disability Outcomes

Total daily activity was significantly associated with each of the mobility disability outcomes. Each 100 000 count (~1 SD) of higher total daily activity at the analytic baseline was associated with lower odds of prevalent self-reported mobility disability (odds ratio [OR]: 0.26 [95% CI: .17–0.41]), lower risk of incident self-reported mobility disability (hazard ratio: 0.50 [95% CI: .32–0.78]), and lower risk of loss of ambulation (OR: 0.68 [95% CI: .55–0.85]).

Mobility Abilities, Total Daily Activity, and Mobility Disability Outcomes

For primary analyses, we examined models with significant mobility abilities and TDPA. TDPA remained statistically significant for prevalent self-reported mobility disability, though the odds ratio estimate was attenuated to 0.52 (95% CI: .31–0.87; Table 2). However, when modeled together, TDPA was no longer significantly associated with incident self-reported mobility disability or incident loss of ambulation (Table 2). Further adjusting for visual impairment, BMI, depressive symptoms, BMI, vascular risk factors, and vascular diseases did not change overall results (Supplementary Table S5). Mobility abilities largely remained significantly associated with all 3 measures of mobility disability and their associations were generally not attenuated by the presence of TDPA.

Table 2.

Association Between Mobility Abilities and Mobility-Related Outcomes

		Prevalent Self-Reported Mobility Disability		Incident Self-Reported Mobility Disability		Incident Loss of Ambulation
Mobility Task	Mobility Abilities	OR (95% CI)	p Value	HR (95% CI)	p Value	HR (95% CI)	p Value
32-Foot walk	Regularity	–		–		–
	Cadence	–		–		–
	Pace	–		–		–
	Step time variability	–		–		–
Standing balance	Sway frequency	–		–		–
	Sway jerk	–		–		1.30 (1.10–1.55)	0.002
	Sway magnitude	–		1.26 (1.08–1.47)	0.003	–
TUG walk	Regularity	–		–		–
	Cadence	–		–		–
	Pace	0.68 (0.53–0.89)	0.005	–		0.67 (0.48–0.93)	0.02
	Step time variability	1.29 (1.05–1.59)	0.01	1.21 (1.01–1.44)	0.04	–
TUG turn	Yaw magnitude	0.72 (0.52–1.00)	0.05	0.72 (0.56–0.92)	0.009	0.68 (0.47–1.01)	0.05
TUG stand-up	Complexity	0.67 (0.47–0.94)	0.02	–		–
	Duration	–		1.58 (1.18–2.12)	0.002	–
TUG sit-down	Descension	0.76 (0.58–0.99)	0.04	–		–
	Smooth	1.29 (0.99–1.69)	0.06	–		–
Total daily physical activity*		0.52 (0.31–0.87)	0.01	0.69 (0.43–1.09)	0.11	0.73 (0.41–1.31)	0.29

Notes: Models include all statistically significant motor abilities identified via backwards selection and is adjusted for age, sex, race, and years of education. Models that are further adjusted for vision impairment, depressive symptoms, BMI, vascular risk factors, and vascular diseases are available in the Supplementary Material. HR = hazard ratio; OR = odds ratio; TUG = timed up and go.

*Effect per 100 000 counts (~1 SD) of higher total daily activity.

Because we examined 3 outcomes in our primary analysis, we examined whether our results changed if we applied a stricter threshold for significance for multiple comparisons (0.05/3, p < .017). While some mobility abilities were not statistically significant using this threshold, our main finding was unchanged: TDPA’s association with the different outcomes was attenuated when terms for mobility abilities were included in the same model.

Testing if Mobility Abilities Link TDPA with Mobility Disability Outcomes

As noted above, both mobility abilities and TDPA were strongly associated with mobility disability outcomes. So, the attenuation of the association of TDPA with mobility disability by the addition of terms for mobility abilities in a single model might occur if mobility abilities linked TDPA with mobility disability. Figure 1 shows our hypothesized mediation model with direct effects of TDPA on mobility disability and indirect effects through mobility abilities.

Prevalent self-reported mobility disability

Direct effects in the mediation models for prevalent self-reported mobility disability can be interpreted as the odds of mobility disability associated with 1 SD (100 000 activity counts) of higher total daily activity that can be directly attributed to total daily activity alone. Indirect effects are interpreted as the odds of prevalent self-reported mobility disability associated with 1 SD of higher total daily activity that are mediated through better mobility abilities. When considered together, mobility abilities mediated 53.5% of the effect of total daily activity on lower odds of self-reported mobility disability (p < .001; Table 3).

Table 3.

Effect of Total Daily Activity on Mobility Self-Reported Mobility Disability and Loss of Ambulation Mediated by Mobility Abilities

Disability Outcome	Mobility Abilities	Direct Effect of Total Daily Activity	Indirect Effect of Total Daily Activity Through Motor Ability	Percent of Total Effect Mediated by Motor Ability
		OR (95% CI)	OR (95% CI)	% (p Value)
Prevalent self-reported mobility disability	Combined mobility abilities	0.92 (0.85 to 1.00)	0.90 (0.87 to 0.94)	53.5% (p < .001)
	TUG complexity	0.85 (0.80 to 0.92)	0.98 (0.96 to 1.00)	11.0% (< 0.001)
	TUG descension	0.85 (0.80 to 0.91)	0.98 (0.96 to 1.0)	10.1% (< 0.001)
	TUG smooth	0.83 (0.78 to 0.89)	1.00 (1.00 to 1.01)	NS*
	TUG pace	0.88 (0.83 to 0.95)	0.94 (0.91 to 0.96)	32.4% (< 0.001)
	TUG step time variability	0.84 (0.79 to 0.90)	0.99 (0.97 to 1.01)	5.6% (< 0.001)
	TUG turn yaw magnitude	0.90 (0.85 to 1.00)	0.93 (0.90 to 0.96)	40.2% (< 0.001)
		Coef. (95% CI)†	Coef. (95% CI)†	% (p Value)
Incident self-reported mobility disability	Combined mobility abilities	1.06 (−0.82 to 2.26)	0.82 (0.22 to 1.59)	40.3% (0.03)
	Standing balance sway magnitude	2.32 (1.20 to 3.15)	−0.09 (−0.38 to 0.12)	NS*
	TUG step time variability	2.27 (1.180 to 3.08)	0.005 (−0.36 to 0.34)	NS*
	TUG stand-up duration	1.79 (0.33 to 2.76)	0.32 (0.03 to 0.78)	13.5% (0.04)
	TUG turn yaw magnitude	1.36 (−0.39 to 2.49)	0.66 (0.17 to 1.32)	30.3% (0.01)
		Coef. (95% CI)†	Coef. (95% CI)†	% (p Value)
Incident loss of ambulation	Combined mobility abilities	5.16 (6.76 to 12.59)	6.44 (2.57 to 13.22)	50.0% (0.02)
	Standing balance sway jerk	12.60 (4.45 to 20.57)	−0.37 (−1.92 to0.77)	NS*
	TUG pace	6.09 (−4.42 to 12.89)	5.19 (1.97 to 10.85)	41.6% (0.01)
	TUG turn yaw magnitude	5.34 (−6.40 to 12.47)	6.29 (2.28 to 13.33)	48.8% (0.01)

Notes: All models were adjusted for age, sex, race, years of education, vision impairment, depressive symptoms, BMI, vascular risk factors, and vascular diseases. OR = odds ratio; TUG = timed up and go.

^*Mediation percentage not statistically significant.

^†Coefficient from survival regression models (24). Positive effects can be interpreted as the number of years that incident self-reported mobility disability or incident loss of ambulation is postponed by 1 SD (100 000 activity counts) increase in total daily activity.

When examining mobility abilities separately, there was evidence of an indirect effect of total daily activity through mobility abilities on the odds of prevalent self-reported mobility disability for each TUG complex, sit-down, smooth, walking pace, step time variability, and yaw magnitude (p < .001 for each; Table 3), by which higher total daily activity was associated with lower odds of self-reported mobility disability due to better motor ability.

Except for TUG smoothness, the percent of the effect of total daily activity mediated through mobility abilities was statistically significant and ranged from 5.6% (TUG step time variability) to 40.2% (TUG yaw magnitude; p < .001 for each).

Incident self-reported mobility disability

For mediation models examining incident self-reported mobility disability, coefficients from direct effects from these survival analyses can be interpreted as the average number of years per 100 000 counts of higher total daily activity that self-reported mobility disability is postponed (for positive coefficients; negative coefficients denote earlier mobility disability) that is directly attributed to total daily activity. Indirect effects indicate the years of mobility disability postponed per 100 000 counts of higher total daily activity through better motor ability.

Combined, the mobility abilities mediated 40.3% of the effect of total daily activity on lower risk of incident self-reported mobility disability (Table 3). Separately, there was evidence of an indirect effect of total daily activity through mobility abilities on the time-to-incident self-reported mobility disability for each standing balance sway magnitude as well as TUG step time variability, stand-up duration, and yaw magnitude (Table 3). However, the percentage of the effect mediated through mobility abilities was only statistically significant for the TUG stand-up duration and yaw magnitude.

Incident loss of ambulation

For incident loss of ambulation, combined, mobility abilities mediated 50.0% (p = .02; Table 3) of the effect of total daily activity on incident loss of ambulation. There was evidence of an indirect effect of total daily activity through mobility abilities for standing balance sway jerk, though the percentage of the effect mediated was not statistically significant. TUG pace mediated 41.6% of the effect of total daily activity (p = .01; Table 3), while TUG yaw magnitude mediated 48.8% of the effect (p = .01).

Discussion

In this study of more than 700 community-dwelling older adults, we found unique combinations of sensor-derived mobility abilities that were associated with varied mobility disability phenotypes, with only turning yaw magnitude being associated with each. Associations of TDPA and incident, but not prevalent, mobility disability phenotypes were attenuated when considered with mobility abilities. Novel mediation models using these combinations of mobility abilities suggest that mobility abilities mediate up to one-half of the association of TDPA with different phenotypes of mobility disability. These findings underscore the potential utility for using unobtrusive sensors to quantify a broader range of facets of gait and balance in older adults and may inform on the design of intervention studies and targeted therapies to maintain ambulation and reduce mobility disability in late life.

The current study extends prior studies and adds to the literature in several important ways. Our use of sensors for capturing both TDPA and mobility abilities is unique and circumvents the limitations of prior studies that employed quantitative measures of mobility abilities but self-report physical activity measures (25). It also extends our prior work that found that both sensor-derived mobility abilities and TDPA alone are associated with mobility disability, but did not examine these factors together (10). Since our prior work showed that mobility and TDPA are weakly correlated, we hypothesized that both would remain associated with mobility disability when included in a single model. However, our primary analyses did not support our hypothesis, but rather showed that including terms for mobility abilities attenuated the association of TDPA with mobility disability. Based on the large number of prior physical activity intervention studies designed to reduce mobility disability, this led us to examine if the attenuation of TDPA seen for incident disability outcomes was due to the fact that mobility abilities might link (ie, mediate) TDPA with mobility disability (26–30). A novel aspect of our mediation analysis was that we examined the mediation effects of multiple mobility abilities together and individually for each of the mobility disability phenotypic outcomes. These phenotypes vary in severity with loss of ambulation being the most severe, though the percent of the effect of TDPA mediated by mobility abilities for each outcome was similar, as a subset of mobility abilities mediated about 40%–50% of the effect of TDPA When considered together, the percentage of the effect mediated through combined mobility abilities is higher than any single ability, demonstrating the potential utility of analyzing these metrics together. TDPA still had a sizeable direct effect that was not due to mobility abilities, highlighting the varied pathways through which higher TDPA may contribute to the reduction of mobility disability in aging adults.

Furthermore, by using detailed measures of sensor-derived mobility abilities, we were able to extend our prior studies showing that different combinations of sensor-derived mobility abilities are associated with different mobility disabilities (7). TUG turn yaw magnitude was the only mobility ability that was independently associated with all 3 mobility disability outcomes. Turning requires more cognitive and motor resources as compared with straight walking as both legs move asynchronously at different speeds while the body’s center of gravity, balance, and posture must be maintained to prevent falls (31–33). The yaw magnitude is a measure of turning efficiency and axial rotation and is not captured in traditional TUG testing that only measures the time to complete TUG tasks (31). Previously, we also showed that yaw magnitude is associated with adverse outcomes in late life, including incident parkinsonism (34), disability in activities of daily living (7), and Alzheimer’s dementia (7), even in adults without known neurological diseases. Taken together the current results showing that yaw magnitude significantly mediated a large portion of the association between TDPA and each of the 3 mobility disability phenotypes (approximately 30%–50%), turning efficiency may be a particularly important clinical biomarker for risk stratification. Future studies should investigate whether turning efficiency can be improved through targeted interventions to reduce disability and other adverse health outcomes in aging adults.

Despite being a robust predictor of adverse outcomes, including mobility disability (1), pace (eg, gait speed) from the 32-foot walk was associated with mobility outcomes when examining mobility abilities by subtask for prevalent self-reported mobility disability, but not when included with mobility abilities from other mobility tasks, or for other outcomes. Instead, pace from the TUG walk was independently associated with prevalent self-reported mobility disability and incident loss of ambulation, and it mediated approximately 30%–40% of the association between TDPA and these outcomes. Though pace from the 32-foot walk and the TUG are strongly correlated, it is possible that the task constraints imposed by the TUG that require the sequential linkage of several different movements for its successful completion may account for the associations of TUG pace rather than 32-foot walk pace. The sequential linkage of several TUG subtasks may also account in part for the strong association of TUG yaw magnitude with mobility disability. This idea may be analogous to prior reports that have suggested that cognitive TUG testing may increase the identification of incipient cognitive deficits when an individual is challenged with divided attention (35,36). Further modifications of the instructions (eg, dual task) or subtask complexity assessed as part of the TUG might increase its specificity and sensitivity as a clinical biomarker for the prediction of specific mobility impairments and disability phenotypes.

There are several strengths to this study. First, the inclusion of total daily activity measures obtained with a wrist sensor circumvents participant recall biases. Second, using unobtrusive sensors provided a wider range of mobility abilities that cannot be assessed via conventional gait testing. This is particularly important as one of the most robust mediators was yaw magnitude from the TUG, which requires measures of angular velocity for its quantification. Though the use of sensors to measure mobility and physical activity requires additional computational and analytical resources, this did not overburden our older community-dwelling participants and were easily added to a testing battery performed in participants’ homes. Finally, the use of multiple mediation rather than only focusing on single mediators helped us to better understand the individual and joint contribution of mobility abilities on mobility disability outcomes.

This study has limitations. The cohort is predominantly White, and most participants are female, limiting generalizability. Despite our mediation analyses, this is an observational study that informs but cannot replace further intervention studies. Without longitudinal mobility metrics and actigraphy, we are unable to elucidate the timing of changes in mobility abilities and total daily activity that may provide additional insight into the causal pathway leading to mobility disability. This should be tested in future work. The sensor-derived mobility metrics identified as mediating the association between total daily activity and mobility disability were obtained at a single supervised testing session. Thus, this work is best conceptualized as an initial step that highlights the potential benefits of using combinations of sensors to capture varied motor phenotypes during everyday living.

In conclusion, sensor-derived mobility metrics may distinguish different aspects of mobility disability, linking a more active lifestyle with reduced disability. Findings may inform the design of intervention studies to develop targeted therapies to maintain ambulation and independent living in late life.

Funding

This work was funded by grants from the National Institute on Aging (R01AG56352, R01AG79133, and R01AG15819). Data and resources from the Rush Memory and Aging Project can be requested at https://www.radc.rush.edu. The funders had no role in the design of the study, analyses, interpretation of the data, or decision to submit results for publication.

Conflict of Interest

None.

Author Contributions

B.L.-M. and A.S.B. conceptualized the study. B.L.-M. wrote the original draft. T.W. conducted statistical analyses. A.S.B. provided supervision and mentorship. All authors contributed to data interpretation in reviewing and editing the written manuscript.