Longitudinal Associations Between Walking Speed and Amount of Self-reported Time Spent Walking Over a 9-Year Period in Older Women and Men

Abstract

Background

Age-related limitations in mobility and decreased physical activity appear to be linked cross-sectionally; however, large-scale, longitudinal analyses of the associations between age-related changes in mobility and engagement in physical activity are lacking. In this longitudinal study, we hypothesized that early mobility limitations would contribute to later decreases in physical activity to a larger degree than the reciprocal association of early decreases in physical activity to later mobility limitations.

Methods

Participants were 2,876 initially well-functioning community-dwelling older adults (aged 70–79 years at baseline; 52% women; 39% black) studied over a 9-year period. Usual walking speed and self-reported physical activity (based on minutes per week of walking) were assessed at Years 0 (ie, baseline), 4, and 9. A cross-lagged, longitudinal model assessed the bidirectional associations between walking speed and physical activity over time.

Results

Early change in walking speed between Years 0 and 4 predicted late change in physical activity between Years 4 and 9 (β = .13 p < .001). However, early change in physical activity did not predict late change in walking speed (β = −.01, p = .79). The difference between these two predictive associations was highly significant (p < .001). Associations were independent of baseline demographic and physical health variables, as well as longitudinal changes in grip and quadriceps strength.

Conclusions

The results suggest declining walking speed as a precursor to declining engagement in physical activity, but the converse association was not evident. Improving walking speed may be a method to increase physical activity among elderly individuals.

Slow walking is a common and serious problem of older age. Epidemiological studies of older adults show that walking more slowly predicts development of disability (1,2), cognitive impairment (3,4), and death (5,6). Hence, slow walking in older adults has significant consequences for the individual, the society and the public health system, and, thus, deserves greater attention.

A person who begins to walk more slowly may also walk less overall, resulting in less engagement in physical activity. The associations are likely bidirectional, and the interplay between reduced walking speed and physical activity could amplify disability by causing deterioration in multiple physiological domains. Evidence suggests that greater sedentary behavior is associated with reduced cardiovascular health (7). The disabling effects of moving less and less well can also translate to greater cognitive impairment, faster decline in cognitive function (8), and impaired mood, for example, through reduced social interaction and increased isolation (9).

Despite the assumed link between decline in walking speed and decline in physical activity engagement, little is known about the potentially dynamic, bidirectional relationship between walking speed and physical activity over time in elderly individuals. Specifically, little is known on whether declines in walking speed lead to reduced physical activity levels over time or vice versa. A recent analysis of the InCHIANTI study (10) began to address this topic by determining whether changes in physical activity from baseline to Year 3 predicted physical functioning decline, including walking speed, from baseline to Year 9 among older women and men. The authors observed that increased physical activity or maintenance of a high level of physical activity was associated with maintained physical functioning, whereas decreased physical activity or persistently low physical activity was associated with physical functioning decline. Whether these associations were driven by the overlap during the first 3 years of the study in the physical activity and functioning assessments or whether early changes in physical functioning would predict longer-term changes in physical activity were not addressed.

The current study examined changes in walking speed and the amount of time spent walking (as a proxy for overall levels of physical activity) in a large, biracial cohort of initially well-function older women and men followed over 9 years. A cross-lagged longitudinal model was used to determine whether early changes in walking speed contributed to future changes in amount of time spent walking or, conversely, whether early changes in amount of time spent walking contributed to future changes in walking speed.

Methods and Measures

Study Design and Participants

Participants were drawn from the Health, Aging and Body Composition study, a prospective cohort study of community-dwelling adults aged 70–79 years at baseline, living in either Pittsburgh or Memphis, USA. Participants were recruited from a random sample of Medicare-eligible adults living within the designated zip codes and were eligible if they reported no difficulties performing activities of daily living, walking a quarter mile, or climbing 10 steps without resting. They also had to be free of life-threatening cancers and plan to remain within the study area for at least 3 years. Men and black individuals were oversampled. Baseline assessments were completed between May 1997 and June 1998, and participants were followed with yearly or semiannual in-person visits. The study was approved by the institutional review boards at the University of Pittsburgh, the University of Tennessee Memphis, and the University of California–San Francisco. All participants gave written informed consent.

Longitudinal Outcome Variables

To examine early and late changes in time spent walking and walking speed, we utilized assessments from baseline, Year 4, and Year 9 of the Health, Aging and Body Composition study. This allowed for early and late change to be roughly equal periods of time. Self-reported time spent walking (minute per week) was based on the previous week using a standardized questionnaire (11). Participants were first asked whether they had engaged in walking at least 10 times in the past 12 months and then were asked had they engaged in walking at all in the past 7 days. Those answering in the affirmative were asked about the total number of minutes of walking over the past 7 days. Walking included for exercise and any other type (eg, transportation). The outcome variable was minutes per week of walking; individuals who had not engaged in walking at least 10 times in the past 12 months or had not engaged in walking in the past 7 days received a score of zero. We did not observe a linear or quadratic association between time spent walking scores and the time of year of the assessment (all R² ≤ .01).

Walking speed (meter per second) at usual pace is a valid and reliable marker of physical performance in older adults (1) and was assessed over a 6-m walkway (Years 0 and 9) and over a 20-m walkway (Year 4). Participants were instructed: “Place your feet with your toes behind, but touching the starting line. Wait until I say ‘GO’. Remember, I want you to walk at a comfortable pace.” Participants began at a standing position and timing was initiated with the first footfall over the starting line. Timing was stopped with the first footfall across or touching the finish line. Similar to previous research (2,5), 20-m walking speed was converted to 6-m walking speed using a conversion formula derived using data from 1,342 individuals who completed both walking speed assessments at Year 9:

Estimated 6-m walking speed = 0.171 + 0.834 × (20-m walking speed).

Descriptive Variables and Covariates

Potential confounding variables included the following variables assessed at baseline: demographics, body mass index (kilogram per square meter), self-reported chronic disease conditions and health behavior, and study site. Demographics included age, sex, race, and educational attainment. Health behavior included smoking status (current, former, or never) and frequency of alcohol consumption. Self-reported chronic disease conditions included cerebrovascular disease, diabetes, and coronary heart disease. Diabetes was confirmed by medication use. Executive functioning and information processing were measured using the Digit Symbol Substitution Test, which consists of a series of numbers (1–9) and corresponding symbols. Participants are requested to draw the correct symbols for given digits during a 90-second time period (12). General cognitive functioning was measured using the Modified Mini-Mental Status Examination (3MS). The 3MS is a comprehensive test of orientation, attention, calculation, language, and short-term memory (13). Additionally, we included measures of grip and quadriceps strength, which were measured repeatedly from baseline to Year 9. Quadriceps strength was measured by knee extension using an isokinetic Kin-Com dynamometer (model 125 AP; Chattanooga, TN). The right leg was tested (unless injured or affected by a condition) at 60° per second, three times, in the concentric mode. Grip strength was measuring using an isometric dynamometer (Jamar, Bolingbrook, IL), while sitting with the arm resting on a table and elbow at approximately a 90° angle. Two trials were completed for each hand, and the average of the four trials was computed; handle spacing was adjusted for each participant. For grip strength, we included data from the same time points as for the primary outcome variables (ie, baseline, Year 4, and Year 9). For quadriceps strength, Year 9 data were available for only ~10% of the sample; therefore, we used data from baseline, Year 4, and Year 6.

Statistical Analyses

Preliminary analyses involved visualization of data distributions for departures from normality; to reduce substantial positive skew, self-reported time spent walking scores were transformed as follows: log₁₀ (time spent walking + 1). Next, the longitudinal, bidirectional associations between changes in walking speed and physical activity were analyzed using cross-lagged latent change models within the statistical package Mplus 7.3 (14). This analytic technique allowed us to simultaneously examine whether baseline and early changes in walking speed predicted later changes in physical activity and/or whether baseline and early changes in physical activity predicted later changes in walking speed. All models used maximum likelihood estimations with robust standard errors using the estimator “MLR.” Our primary models included all individuals with baseline data (N = 2,876); missing follow-up data were implicitly imputed uses the maximum likelihood estimator. Several additional models were constructed to determine whether different ways of dealing with missing data substantially altered the findings. This was done because study completers differed on several of the baseline demographic and performance measures compared with study noncompleters (see Supplementary Table 1 for detailed analysis). First, we used Markov chain Monte Carlo simulation to multiply impute missing data (15); the imputation process was limited to the sample known to be alive at Year 9 (n = 2,084) and used all study variables and covariates to create 40 imputed data sets; parameter estimates and standard errors were pooled across the 40 data sets. Second, we restricted the study sample to those participants with complete data at baseline and Year 4 (n = 2,143). Third, we restricted the study sample to those participants with complete data at all three time points (n = 1,324). Fourth and fifth, we only included individuals with a baseline walking speed of ≥1.0 m/s (n = 2,290) or with a baseline 3MS score greater than 90 (n = 1,697) to determine whether the exclusion of individuals with mobility limitations (16) or cognitive impairment, respectively, might affect the results.

Model estimates were evaluated without covariates and then after adjusting for the following baseline variables: study site, education, race, baseline age, body mass index, sex, baseline smoking and drinking status, and prevalent diabetes, cardiovascular disease, and cerebrovascular disease. Two additional models were constructed to determine whether the associations observed between walking speed and physical activity were independent of physical strength: one controlling for changes in grip strength and the second controlling for changes in quadriceps strength. For the primary analyses, we report standardized estimates (β) and p values. Significance was set at a Bonferroni-corrected p < .0056 to account for the nine cross-domain estimates within the model (α/n = .05/9 = .0056). We used the model test feature within Mplus to directly test whether the predictive association between walking speed and future physical activity was statistically different from the equivalent association between physical activity and future walking speed.

Results

The study sample consisted of 2,876 individuals (94% of the total baseline participants) with baseline data on the measures of interest (Table 1). Participants ranged in age from 69–79 years old. Roughly half were women, 39% were black, 44% had education above high school, and less than 20% had chronic disease conditions. The boxplots in Figure 1 summarize changes in walking speed (Figure 1A) and time spent walking (Figure 1B) over the course of the study period. The range of scores is shown using only the observed data available at each time point (darker boxplots) and with multiple imputation of missing data (lighter boxplots).

Table 1.

Characteristics of the Study Sample

Variable	Number of Observations Available for Variable	Mean (SD) or n (%)
Age, y	2,876	73.6 (2.9)
Sex, women	2,876	1,499 (52%)
Race, black	2,876	1,134 (39%)
Education, > high school	2,869	1,247 (44%)
Digit symbol substitution test	2,876	36.6 (13.5)
Modified Mini-Mental State Examination	2,876
Median (25th, 75th percentiles)		92 (87, 96)
>90 score		1,697 (59%)
Body mass index (kg/m2)	2,876	27.4 (4.8)
Chronic disease conditions
Coronary heart disease	2,876	482 (17%)
Cerebrovascular disease	2,876	208 (7%)
Diabetes	2,876	426 (15%)
Health behavior
Current or former smoker	2,872	1,619 (56%)
Alcohol consumption, ≥ once per week	2,867	854 (30%)
Study site, Pittsburgh	2,876	1467 (51%)
Self-reported walking (min/wk)
Baseline	2,876	125.3 (239.9)
Year 4	2,457	83.9 (156.2)
Year 9	1,769	68.9 (135.9)
Usual walking speed (m/s)
Baseline	2,876	1.18 (.23)
Year 4	2,159	1.09 (.18)
Year 9	1,423	1.01 (.25)
Grip strength (mean kg)
Baseline	2,744	29.8 (10.1)
Year 4	2,214	28.6 (9.5)
Year 9	1,503	25.4 (8.8)
Quadriceps strength (mean Nm)
Baseline	2,496	106.3 (38.4)
Year 4	1,992	95.1 (35.1)
Year 6	1,854	89.2 (32.9)
Deceased at Year 4	2,876	236 (8%)
Deceased at Year 9	2,876	780 (27%)

Figure 1.

Boxplots of walking speed (A) and time spent walking (B) scores at each time point. Note that time spent walking scores have been log-transformed. Each box represents the interquartile range; horizontal line within each box is the median score; the whiskers (vertical lines extending from the box) cover 1.5× interquartile range, and values beyond this are represented by individual filled circles. Boxplots were prepared using the observed data at each point and with missing data imputed.

Table 2 provides the cross-domain estimates from the cross-lagged latent change models with the exclusion and inclusion of covariates; the covariate-adjusted model is depicted in Figure 2. Here we focus on the results from the covariate-adjusted model. Baseline walking speed was a significant predictor of early (β = .09, p < .001) and late (β = .17, p < .001) change in physical activity. On the other hand, baseline physical activity predicted early (β = .05, p = .001) but not late (β = .002, p = .94) change in walking speed. Moreover, early change in walking speed predicted late change in physical activity (β = .13, p < .001); however, early change in physical activity did not predict late change in walking speed (β = −.01, p = .79). The association between early change in walking speed and late change in physical activity was significantly stronger from the association between early change in physical activity and late change in walking speed (Wald test [df = 1] = 12.74, p < .001). Beyond these predictive associations, we observed that concurrent change in walking speed and physical activity were correlated, both during the early (β = .11, p < .001) and late (β = .08, p < .001) phases of the study.

Table 2.

Standardized Associations Between Walking Speed and Physical Activity

Model Paths	Unadjusted	Adjusted for Covariatesa
Predictive associations
BL walking speed → early Δwalking amount	.15 (.02)*	.09 (.02)*
BL walking amount → early Δwalking speed	.09 (.02)*	.05 (.02)*
BL walking speed → later Δwalking amount	.22 (.03)*	.17 (.03)*
BL walking amount → later Δwalking speed	.02 (.02)	.002 (.03)
Early Δwalking speed → later Δwalking amount	.15 (.03)*	.13 (.03)*
Early Δwalking amount → later Δwalking speed	.01 (.02)	−.01 (.02)
Residual correlations
BL walking speed ↔ BL walking amount	.19 (.02)*	.08 (.02)*
Early Δwalking speed ↔ early Δwalking amount	.14 (.02)*	.11 (.02)*
Late Δwalking speed ↔ late Δwalking amount	.09 (.02)*	.08 (.02)*

Note: BL = baseline; PA = physical activity as indexed by self-reported walking (min/wk). Standardized estimates (and standard errors) are shown.

^aCovariates include clinical site, education, age, race, baseline body mass index, gender, baseline smoking and drinking status, and prevalent diabetes, cardiovascular disease, and cerebrovascular disease.

*p < .0056.

Figure 2.

The depiction of the main findings of the study. Standardized estimates (and standard errors) are provided for cross-domain associations. Gray lines are within-domain associations and are not of interest in the current study. To simplify the model, covariates are not shown but included clinical site, education, age, race, baseline body mass index, gender, baseline smoking and drinking status, and prevalent diabetes, cardiovascular disease, and cerebrovascular disease. Further adjustment for changes in grip and quadriceps strength did not substantially alter these associations. ε = residual change in the latent variable not accounted for covariates or previous assessments. Solid lines with bolded estimates indicate associations that are significant based on Bonferroni correction (p < .0056).

The associations between the covariates and the outcome variables estimated from the covariate-adjusted model are provided in Supplementary Table 2. Of note, higher baseline age predicted greater declines in walking speed from baseline to Year 4 (β = −.19, p < .001) and from Year 5 to Year 9 (β = −.14, p < .001). Individuals with higher education showed smaller reductions in walking speed from baseline to Year 4 (β = .11, p < .001). Women showed greater declines from baseline to Year 4 in physical activity (β = −.14, p < .001) and walking speed (β = −.20, p < .001). Finally, individuals with higher body mass index showed greater declines from baseline to Year 4 in physical activity (β = −.07, p < .001) and walking speed (β = −.15, p < .001).

Table 3 provides the cross-domain estimates between walking speed and physical activity after adjusting for changes in grip and quadriceps strength in addition to the covariates from the previous model. As shown, changes in physical strength did not alter the associations between walking speed and physical activity, with early changes in walking speed remaining a significant predictor of later changes in walking amount (β = .12, p < .001 when adjusting for grip strength and β = .13, p < .001 when adjusting for quadriceps strength).

Table 3.

Standardized Associations Between Walking Speed and Physical Activity After Adjusting for Grip and Quadriceps Strength

Model Paths	Adjusted for Grip Strengtha	Adjusted for Quad Strengtha
Predictive associations
BL walking speed → early Δwalking amount	.08 (.02)*	.08 (.02)*
BL walking amount → early Δwalking speed	.05 (.02)*	.05 (.02)*
BL walking speed → later Δwalking amount	.16 (.03)*	.18 (.03)*
BL walking amount → later Δwalking speed	.001 (.03)	.004 (.03)
Early Δwalking speed → later Δwalking amount	.12 (.03)*	.13 (.03)*
Early Δwalking amount → later Δwalking speed	−.01 (.03)	−.01 (.02)
Residual correlations
BL walking speed ↔ BL walking amount	.09 (.02)*	.08 (.02)*
Early Δwalking speed ↔ early Δwalking amount	.11 (.02)	.11 (.02)*
Late Δwalking speed ↔ late Δwalking amount	.07 (.02)*	.08 (.02)*

Note: BL = baseline; PA = physical activity as indexed by self-reported walking (min/wk). Standardized estimates (and standard errors) are shown.

^aCovariates also include clinical site, education, age, race, baseline body mass index, gender, baseline smoking and drinking status, and prevalent diabetes, cardiovascular disease, and cerebrovascular disease.

*p < .0056.

Across all sensitivity analyses, which addressed missing data in various ways (see Supplementary Tables 3–7), the standardized estimates were similar to those in the primary analyses reported above. Specifically, the association between early change in walking speed and late change in physical activity ranged from β = .08 to .14 in the adjusted analyses; the association between early change in physical activity and late change in walking speed ranged from β = −.01 to .03.

Discussion

This longitudinal study of older adults followed over a 9-year period showed that slower baseline walking speed and early declines in walking speed over the first 4 years of the study predicted decline in physical activity levels over the second 5 years of the study. In contrast, neither baseline physical activity nor early changes in physical activity levels predicted later changes in walking speed. These findings support the notion that an individual who begins to walk more slowly is more likely to subsequently walk less overall, potentially resulting in less engagement in physical activity. Notably, these results were similar in analyses controlling for changes in physical strength—either lower or upper body strength—which suggests that these associations implicate walking speed specifically rather than physical strength more generally.

Reduced walking speed may set off a negative cascade of events leading to disability, which might occur in part through reduced physical activity engagement (17). Regular physical activity has myriad benefits for older adults, including the prevention and reduction of chronic diseases (18), promotion of emotional well-being (9), and maintenance of cognitive and brain health (19). As previous physical activity-based interventions have been shown to promote mobility (20,21), decreased physical activity could further exacerbate limited mobility.

One way to better understand the bidirectional and dynamic association between mobility and physical activity is to add physical activity outcomes to intervention studies for mobility, as well as the inclusion of mobility outcomes to intervention studies for physical activity. Such studies would answer questions of causality that cannot be addressed satisfactorily in observational studies. A recent large-scale intervention—the LIFE study—demonstrated that moderate-intensity, multicomponent physical activity promoted physical activity behavior and reduced physical disability (20), and promoted walking speed over 400 m but not over 4 m (22). A previous meta-analysis found that high-intensity, but not moderate- or low-intensity, exercise promoted faster walking speed (21). Most of the previous interventions have focused on improving strength and endurance without much attention on improving the motor control of walking (23). A recent study showed that motor learning exercises lead to greater improvements in walking speed and other mobility outcomes compared with a standard exercise program with endurance and strength training components (23,24). An intriguing pilot study provided suggestive, though not statistically significant, evidence that motor sequence learning to improve gait lead to greater maintenance of physical activity over time in comparison to standard rehabilitation program of strength and balance training among individuals with mobility limitations (25). Larger future studies are needed to confirm this tentative relationship between gait improvement and physical activity engagement.

Unlike a recent study of the InCHIANTI cohort (10), we did not observe that early changes in walking predicted subsequent changes in walking speed; however, there are some notable differences in these studies that might explain the discrepancy. Most importantly, in the current study, early and later change were temporally nonoverlapping, whereas in this previous study, the first 3 years of the study were included in both the assessment of early change in physical activity and long-term changes in physical functioning. Thus, it is unknown the degree to which a correlation in concurrent change contributed to the effects observed. Indeed, in the current study, we did observe that early change in walking was correlated with concurrent change in walking speed.

The complex nature of limited mobility among older adults needs to be taken into consideration and investigated closely as we investigate this association. Namely, the sheer number of disorders, diseases, and conditions that may cause mobility limitations in older age is so large that it adds a complexity to understanding the true nature of this relationship. Furthermore, some diseases of aging progress slowly without overt clinical symptoms until years after underlying pathology has begun.

Several strengths of this study should be highlighted. First, this study has complete data on many covariates, including a large number of self-reported chronic disease conditions, health behaviors, and medication use, and the associations remained robust to adjustments for all of these factors. Furthermore, the large size of the sample allowed us to investigate this association while treating missing data in a variety of ways (from using only those individuals with complete data across all time points to multiply imputing missing data), and we observed the robust nature of the association in all scenarios.

The current study has several limitations, which should be taken into consideration. Physical activity was measured via questionnaires and was self-reported time spent walking, so it does not include a wide range of activity. Thus, it is unclear whether the associations observed herein would generalize to overall levels of physical activity. For example, early changes in walking speed may not predict subsequent decline in other forms of physical activity. This limitation notwithstanding, among older adults, walking is the most common physical activity (26) and was measured repeatedly over time in this cohort, which was required for the chosen analyses. Self-report measures of physical activity may also be affected by recall bias and response bias. These biases can lead individuals to over-report their levels of activity (27) and others to under-report their levels of activity (28). Furthermore, our questionnaire does not provide information about the intensity of the walking, which might be an important moderator of the association between time spent walking and walking speed. It is possible that the use of objective physical activity measurement—which in turn would allow for an analysis of activity intensity—might lead to different conclusions than those herein, so future studies should assess if objectively measured physical activity and mobility have similar associations. Another limitation is that the observational nature of this study precludes us from concluding that decline in walking speed is causing reductions in physical activity engagement. Finally, it should be noted that although statistical significance was set at a conservative level using Bonferroni correction, the size of the significant associations is considered modest based on the convention that standardized estimates of .10 are small and .30 are moderate (29). However, the meaningfulness of the observed effect sizes should also be considered in the context of other variables. Specifically, the size for the association between early change in walking speed and later change in physical activity (absolute value = .13) was similar to the size of the effect of chronological age on later change in physical activity (absolute value = .14). The similarity in effect sizes suggests an important role of walking speed decline.

Conclusions

This longitudinal analysis of walking speed and amount of time spent walking suggests a directionality in the association, such that reduced walking speed over time predicted future reductions in amount of time spent walking. The reciprocal association (ie, reduced walking time did not lead to reduced walking speed) was not evident. In light of the important role of physical activity engagement in myriad aspects of well-being and health, the current findings might provide insight into why slow walking can have serious consequences, including disability and death. Thus, greater attention should be paid to age-related decreases in walking speed, including methods to rehabilitate slow walking speed.

Supplementary Material

Supplementary data is available at The Journals of Gerontology, Series A: Biological Sciences and Medical Sciences online.

Funding

This research was supported by National Institute on Aging (NIA) Contracts N01-AG-6-2101, N01-AG-6-2103, N01-AG-6-2106; NIA grant R01-AG028050; and NINR grant R01-NR012459. This research was supported in part by the Intramural Research Program of the NIH, National Institute on Aging.

Conflict of Interest

None reported.