TURNING SPEED AS A MORE RESPONSIVE METRIC OF AGE-RELATED DECLINE IN MOBILITY: A COMPARATIVE STUDY WITH GAIT SPEED

Abstract

Background

Navigating your environment requires both straight-line gait as well as turning. Gait speed normative values are well established and utilized in determining a person’s functional status, however, it has limitations. This study sought to examine whether turning speed declines with age and how it compared to gait speed age-related decline.

Methods

A secondary analysis was performed on 275 community dwelling adults between the ages of 18–88 that performed a timed walking test with an inertial measurement unit on their lumbar spine. Turning speed and walking speed were extracted for each participant. A series of mixed models were compared, and Akaike’s Information Criterion was used to determine the best fit model between age and turning speed and age and gait speed.

Findings

Turning speed and gait speed normative values were reported for each age decade. A linear model with a random intercept of “Condition” was used to assess the relationship between age and turning speed. The results indicated a significant negative relationship between age and turning speed (B=−0.66, p<0.001). A spline-fit model determined a significant negative relationship between age and gait speed after the age of 65 (B=−0.0097, p=0.002). The effect of age on gait speed before age 65 was not significant.

Interpretation

Turning speed significantly declines with age in a linear fashion while gait speed begins to decline after age 65. Turning speed may be more responsive to age than gait speed. More research is needed to determine if the decline in turning speed with age is associated with a decline in function.

1. Introduction

Being able to perform activities of daily living (ADLs) while aging is directly linked to independent living, as physicians and adult care social workers use ADLs to determine if a person needs assisted living. The ability to walk independently and turn safely while transferring are crucial. Independent community mobility via ambulation requires individuals to perform straight-line gait as well as turning. Gait speed has been used to characterize performance of straight line gait, with gait speed being described as the 6^th vital sign such that a decline in gait speed indicates an increase in morbidity and mortality¹. It has value in providing a snapshot of a patient’s mobility disability ^2,3 and cognitive impairment ^4–6 with minimal equipment and requiring minimal training for the test administrator. However, there are limitations to gait speed as a clinical measure. For example, as a fall predictor, gait speed is not the most robust gait variable ⁷ in predicting falls and is affected by variability in testing procedures⁸. With these shortcomings, solely using gait speed as an indicator of functional decline could result in failing to identify an individual at higher risk for adverse events and unnecessary health consequences.

Compared to straight-line gait speed, turning is more complex, requiring coordination of the vestibular, visual and somatosensory systems ^9–11. For these reasons, turning may capture the decline of sensory systems that contribute to functional decline and potentially higher risk for falls earlier than gait speed. Portions of the vestibular system, which process angular motion (i.e., turning), demonstrate age-related degradation in neural components starting around 40 years of age^12,13; however, it is unknown whether this vestibular degradation affects clinical and community performance. Considering that the quality of turns ¹⁴ may be more indicative of fall risk than quantity of turns, appropriate measurement of turning is crucial to make clinical judgments. Peak turning speed has proven to be a metric that differentiates between healthy controls and individuals with subtle postural deficits while still maintaining similar gait speed^14–16.

Clinical walking tests (e.g., 2-minute walk test (2MWT) ¹⁷, require individuals to walk back and forth between two floor markers, performing a U-turn at each marker. Performing these tests while instrumented with inertial measurement units (IMUs) allows for the concurrent measurement of multiple turns and straight-line gait. To our knowledge, there are no studies that have measured average peak turn speed and gait speed captured during timed walking tests in individuals across a broad range of ages with the objective of gaining insight into differences in the association between age and these gait metrics. Examination of the differential effect of age on turning speed and gait speed would provide insights into age-related changes in varied aspects of gait function and document normal range values of peak turning speed.

The present study served two purposes: 1) to examine age-related changes in turning speed to determine if the trajectory of the decline is different from the trajectory of the decline of gait speed 2) to report normative turning speed values for fast pace U-turns, to aid in clinical interpretation as wearable sensor technology continues to advance in clinical settings. We hypothesize that a decline in turning speed will occur earlier in life than the decline in gait speed.

2. Methods

All participants provided written informed consent prior to data collection. Data were collected at two sites, Oregon Health & Science University (OHSU) and the University of Utah (UoU). The participant data for this study was pulled from five larger studies which included a timed walk test (NIH grants: R44AG055388, R01AG006457, VA Merit award: I01 RX001075, National Multiple Sclerosis Society RG-1701–26763). Specific inclusion and exclusion criteria for each study are listed in Supplemental Materials, however this analysis required participants to be healthy independent ambulators, between ages 18 and 90. This study was approved by the University of Utah Institutional Review Board (IRB; Salt Lake City, Utah). The data collected at OHSU was approved by the OHSU IRB (Portland, Oregon).

Participants were fitted with one wearable inertial measurement unit (IMU; Opal, APDM Wearable Technologies, a Clario company, Portland, OR) at approximately L5 of the lumbar spine secured with an elastic belt. IMUs with a triaxial accelerometer, gyroscope and magnetometer were used to record gait and turning metrics of each participant as they performed a walk test (sampling frequency 128 Hz). Such IMUs allowed for the accurate measurement of straight-line gait ^18–20 and turning ^21,22. Depending on the data collection site, the length of the walking course varied between studies (Table 1), however each participant was directed to walk a path between two points, turn around after floor markers and repeat until the time was complete. Participants were instructed (instructions not standardized) to walk as quickly and safely as possible to capture their maximum walking performance.

Table 1.

Study condition and walking test characteristics.

Condition	Duration/length of walk test	Length of walking path	Path markers
1	3 minutes	10 meters	Floor lines
2	2 minutes	7 meters	Floor lines
3	400 meters	20 meters	Cones
4	2 minutes	7.62 meters	Floor lines
5	6 minutes	25 meters	Floor lines
6	2 minutes	25 meters	Floor lines

2.1. Outcomes of Interest and Data Processing

After collection, accelerometer and gyroscope data were exported and processed with custom MATLAB algorithms (Mathworks 2018b, Natick, MA) developed by our laboratory. This filtered the data using a fourth-order Butterworth filter with a 6 Hz cutoff frequency. For trials that were longer than 2 minutes, the file was cropped and only the first 2 minutes were analyzed.

2.2. Turning Speed

To capture peak turning speed in the 2MWT test, peaks were selected from the lumbar gyroscope data that corresponded to 180° turns at the end of the walking path. The researcher manually input the total number of turns (N), and the script found the N largest amplitude peaks in the lumbar gyroscope data. The absolute values of each turn’s speed were calculated, and the overall mean was calculated to represent the average peak speed of trunk turns occurring for each 2MWT trial. The data processing details can be found in the validation study by Paul et al²¹.

2.3. Gait Speed

Data for each turn was also isolated to track the time of turn, and time between turns. With the known path length, walking speed between each turn, along the straight walking path, was calculated as gait speed. The gait speed for each lap was averaged and reported as the average gait speed for the entire trial.

2.4. Statistical Analysis

All statistical analyses were performed using R and R Studio^23–25. Participants were categorized into age groups by decade for descriptive statistics, but age was kept as a continuous variable in regression models. Turning peak speed and gait speed descriptive statistics including mean, standard deviation, percentiles and range were calculated for each age decade. To quantify age-related differences in turning speed and gait speed, we constructed a series of mixed-effect regression models in which these outcomes were regressed onto the predictors of sex, BMI, and age, with a random-intercept of “Condition.” Models were estimated using full maximum-likelihood. The condition random-intercept reflects the test conditions under which the gait variables were extracted (Conditions 1–6). (Supplemental Figure A) For both outcomes, we tested an intercept only model, a linear age model, a quadratic age model, a cubic age model, and model with age and a single knot linear spline. The linear spline model allows the age predictor to have one slope before the knot and then instantaneously change to a new slope after the knot (e.g, immediately changing from linear growth to a plateau). A spline model was included as this model can often provide better fits than polynomial models when non-linear changes are present ²⁶. To find the best fitting knot location, we iteratively fit a series of splines between the ages of 30 and 80. Akaike’s Information Criterion (AIC) was used to select the best fitting model. The AIC penalizes the number of parameters to reduce over-fitting and ensure the generalizability of the model ²⁶. After the best fitting model was chosen, distributional assumptions for the residuals and random-intercepts were checked with quantile-quantile normal plots. Due to violations of normality for gait speed, we estimated robust confidence intervals for all coefficients in all models using a semi-parametric bootstrap ²³. Finally, we used the Welsh-Satterthwaite approximation for the degrees of freedom to obtain p-values for hypothesis tests of the different coefficients (α=0.05).

3. Results

Data from 275 participants (154/118 female/male) were collected for analysis. After visual inspection of the data, three participants had turning velocities over 300°/s. While this turning speed is possible, these data points are three standard deviations above the mean (i.e. >99.8 percentile). With this large of a discrepancy, the three outliers were removed from the rest of the analysis. One participant was removed due to missing demographic data. Demographic information for 271 participants are presented in Table 2 and separated by study name. The majority of data collection for this study occurred at OHSU and between the two sites, OHSU participants were older (OHSU 61.5 ± 14.7, UoU 31.5± 8.7). The average age of the participants was 59.1± 16.4 years. In this sample, across all ages, mean turning speed was 179 ± 34°/s and mean gait speed was 1.13 ± 0.46 m/s. Descriptive statistics such as means, standard deviation, and percentiles are listed in Table 3. Adjusted gait and turning speed by age decade are provided in Supplemental Materials (Table A).

Table 2.

Characteristics of participants grouped by study condition.

Condition	Number of participants (# of female)	Age	Height (cm)	Weight (kg)	BMI
1	39 (24)	60.03 (14.34)	170.30 (10.78)	72.34 (15.98)	24.85 (4.64)
2	54 (53)	63.00 (7.46)	162.60 (7.30)	67.37 (12.16)	25.44 (3.99)
3	72 (35)	53.12 (19.87)	173.00 (10.48)	74.06 (14.11)	24.59 (3.10)
4	84 (32)	68.25 (8.12)	172.10 (9.84)	74.09 (13.09)	24.91 (3.41)
5	7 (3)	30.71 (6.16)	180.90 (14.92)	72.77 (7.80)	22.39 (2.46)
6	15 (7)	32.20 (9.99)	171.90 (9.29)	69.37 (11.82)	23.38 (2.81)

BMI= body mass index; cm= centimeters; kg= kilograms

Table 3.

Descriptive statistics for unadjusted turning speed and gait speed for each decade of age matched gait speed.

Age Category	Number of participants (# of female)	Turning speed °/s							Gait Speed m/s
		Min.	1st Q.	Median	Mean	3rd Q.	Max.	SD	Min.	1st Q.	Median	Mean	3rd Q.	Max.	SD
<30	25 (13)	160.50	177.00	206.10	204.40	223.70	270.00	31.18	0.62	1.24	1.66	1.79	2.41	3.09	0.69
30–39	19 (7)	132.20	161.30	194.40	188.30	213.30	246.10	31.84	0.85	1.19	1.28	1.56	1.53	3.00	0.67
40–49	18 (10)	108.90	157.90	166.90	168.70	183.10	231.80	26.99	0.73	1.01	1.10	1.35	1.49	2.76	0.54
50–59	45 (28)	125.50	153.50	182.20	182.80	206.70	272.00	32.28	0.47	0.97	1.04	1.00	1.22	2.42	0.33
60–69	88 (57)	96.570	156.66	179.64	180.23	206.40	254.38	33.60	0.63	0.80	0.98	1.00	1.13	1.63	0.22
70–79	58 (30)	116.40	143.80	162.90	166.60	182.30	242.80	31.38	0.52	0.77	0.90	0.93	1.06	1.42	0.22
80–89	18 (9)	80.85	139.71	161.51	156.75	176.79	215.58	33.50	0.70	0.74	0.85	0.91	1.02	1.35	0.20

Min= minimum, Q= quarter, Max= maximum, SD= standard deviation, °/s = degrees per second; m/s = meters per second.

Taking into account differences between sites and tests eliminated variability in the raw scores attributed to the conditions, resulting in adjusted values that are more easily comparable. This adjustment ensures a more equitable evaluation of the relationship between gait speed/turning speed and age. Model comparison for turning/gait speed and age are displayed in Supplemental Table B. The best fit model, as determined by the lowest AIC value, for turning speed on age was the linear model (AIC= 2608.6) and the spline model for gait speed on age. For the turning speed spline model (AIC=−176.1), the best fitting knot was at 63 years. However, given that the spline model and the simpler linear model were relatively equivalent (Δ AIC= 0.3), in the interests of parsimony we selected the simpler model. For the gait speed spline model, the best fitting knot was at 65 years.

The results of the mixed-effects model demonstrated that age had a significant negative relationship (B = −0.62, p<0.001) with peak turning speed which was best represented with a linear model. For every one-year increase in age, peak turning speed decreased by 0.62°/s. The spline fit mixed-effects model for gait speed regressed onto age (B= −0.0097, p= 0.002) had a significant negative association. The results revealed that for gait speed, there was no statistically significant effect of age before the knot, suggesting that participants were essentially plateaued from across age groups from 18–65 years old. However, there was a statistically significant change at the knot, indicating that participants started to decline in gait speed after 65 years of age. (Tables 5, Figure 1).

Table 5.

Fixed- and random-effects of the spline model of gait speed.

Random Effects
Group	Variance	SD
Condition (intercept)	0.395	0.628
Residual	0.031	0.176
Fixed Effects
Name	Estimate	CI 2.5%	CI 97.5%	p-value
Intercept	1.395	0.862	1.970	0.001
Male	−0.019	−0.066	0.028	0.446
BMI	−0.004	−0.009	0.002	0.252
Age	0.001	−0.001	0.004	0.360
Spline	−0.010	−0.156	−0.003	0.002

Note there were 271 observations nested within 6 conditions. Male was contrast coded such that Male = +0.5 and Female = −0.5. BMI was centered on the mean BMI. Age was measured continuously in years.

BMI= body mass index; SD= standard deviation; CI= confidence interval.

Figure 1.

Model predictions (lines) and individual observations (points) as a function of Age and Sex for turning speed (Panel A) and gait speed (Panel B). Data are adjusted for the random intercept of condition.

4. Discussion

The goal of this investigation was to determine the relationship between age and turning speed and compare it to the well-studied relationship between age and gait speed. U-turns during timed walking tests were collected with wearable sensors and analyzed in 272 participants between the 18 and 88 years of age. In this analysis, we calculated descriptive turning speed data grouping participants under than 30 years of age together and in each age decade between 30 and 90 years of age. Additionally, we used regression models to determine if age was significantly associated with turning speed and whether there was evidence that turning speed and gait speed differ in the trajectory of age-related changes.

Our main findings showed that turning speed decreased with increasing age. Gait speed also decreased with age. However, this appeared to occur primarily after 65 years of age, based on the improved fit of a regression model that included a single knot linear spline. Thus, gait speed and turning speed appear to have different age-related trajectories and may assess different age-related functions. In particular, turning speed may be helpful as an earlier marker of age-related decline in function, as it appeared to decline linearly across all examined ages, whereas gait speed appeared stable until older age (i.e. >65 years). This supports previous research detecting turning speed differences between young adults and middle aged adults while gait speed remains similar between the groups²⁷.

Based on the results of our model, here is a contextual example of how turning speed differed by age group, the average 80-year-old patient performed U-turns during a walk test at a peak turning speed of 168°/s which was 31°/s slower than a 30-year-old participant. The mechanism underlying decreases in turning speed is not established and is likely multifactorial, to include various potential central and peripheral neurologic and motor changes ^12,28–30. For example, age-related declines in vestibular sensory function are known to advance markedly with advancing age ^30,31.

Regarding the relationship between gait speed and age in our sample, the plateau of gait speed in the younger participants could indicate a ceiling effect. There is a maximum walking speed before one would shift to a running stride and the potential differences in mobility between the age groups are not captured with maximum walking gait speed. Age and gait speed have a significant negative relationship that appears after 65 years. For every one-year increase in age after 65 years, gait speed decreased by 0.009 m/s. This behavior of gait speed across age observed in our data was similar to previous studies ^32–34.

Both models were adjusted for the type of walking task the participant performed. There was not a substantial difference when comparing the unadjusted turning speed model to the adjusted model. However, adjusting gait speed resulted in a noticeable change in the distribution of data for gait speed. Additionally, the standard deviation of the random effect of gait speed relative to the average gait speed (0.63/1.40, Table 5) is much larger than the standard deviation of turning speed to the average (11.38/217.17, Table 4). This suggests that the differences in walking tests appear to have a larger influence for gait speed than they do for turning speed.

Table 4.

Fixed- and random-effects of the linear model of turning speed.

Random Effects
Group	Variance	SD
Condition (intercept)	129.60	11.38
Residual	813.60	28.52
Fixed Effects
Name	Estimate	CI 2.5%	CI 97.5%	p-value
Intercept	217.17	199.86	30.86	<0.001
Male	−9.42	−16.88	−0.80	0.016
BMI	−2.31	−3.32	−1.40	<0.001
Age	−0.62	−0.88	−0.35	<0.001

Note there were 271 observations nested within 6 conditions. Male was contrast coded such that Male = +0.5 and Female = −0.5. BMI was centered on the mean BMI. Age was measured continuously in years.

BMI= body mass index; SD= standard deviation; CI= confidence interval.

In the current study, participants younger than 30 years old performed the fastest average turning speed (210 ±32°/s). The mean peak turning speed for each subsequent age decade reported in our cohorts during a timed walking test was comparable to the turning speed reported by Park and colleagues in their examination of the effect of age on gait variables to include turning during the Instrumented Stand and Walk Test (ISAW)³⁵. Time spent in a turn, turn duration, was negatively correlated with age. Turn duration and peak turn speed are both considered turn quality characteristics and are often positively correlated. However, in the current study, peak turning speed did not have a significant association with age. This could be due to the fact that participants performed only a single turn during the ISAW test for that analysis. Another possibility could be due to the difference in testing instructions. ISAW participants were instructed to walk at their natural pace while the current study participants were instructed to walk as fast and safe as possible.

Gait speed has been identified as the 6^th vital sign ³⁶ and is strongly associated with falls, functional decline and mortality ^6,37,38. Gait speed assessment requires minimal equipment (i.e. stopwatch), is easy to administer and therefore, an accepted clinical assessment. However, gait speed can also be measured with wearable sensors along with turning speed. Turning speed, compared to gait speed, has not been studied as extensively due to the difficulty in quantifying turning without expensive and complex equipment. However, turning research has expanded with the increased use of wearable sensors and has shown turning metrics are more responsive in mobility assessment than gait speed. Capturing turning speed during routine testing such as the 2MWT can add value to functional assessment without a significant increase in burden onto patients or clinicians.

Individuals with known neurological diseases perform turns slower even while still maintaining a similar gait speed compared to healthy controls of similar age ^14–16,39. Relying on gait speed alone as a mobility assessment may misidentify a patient’s functional decline due to aging. From this study, we observed turning speed changed more rapidly with increasing age, compared to gait speed, which is less affected until the 6^th decade of life. As a mobility assessment for patients, established normative values for their age range is important in determining if turning behavior is subpar due to diagnosis-related deficits or due to age. Considering this, and the relationship between turning and falls ^40,41, turning assessments should be included as a part of patients’ mobility measurements to provide a comprehensive mobility assessment. Little research has investigated the improvement of turning after rehabilitation, and in such, turning improvement was determined based on comparison to baseline testing and the control group⁴². As turning assessment becomes more commonplace in the clinic, it is important to consider the effect of age on turning speed to provide an accurate target for rehabilitation. This investigation provides baseline 180° turning metrics for community dwelling adults which can be utilized as a normative range with future turning rehabilitation research in clinical populations.

4.1. Limitations and Directions for Future Research

Despite the relatively large overall sample size, a limitation of this study is the disproportion of participants across each decade. The average turning speed represented may be more accurate with a larger number of participants in each age category. In this study a variety of walking tasks were grouped together and these walking tasks varied in path length, turning markers (floor tape vs cones), test time, and likely visual environment and instruction. The type of turn marker can influence the type of turn a participant executes. Spin turns may be more easily performed with a floor marker versus a cone which requires participant to move around. This variability in the walking tests was accounted for in our models and the models used were run with adjusted data to account for the non-normality of gait speed data which allowed for a better comparison of gait speed and age and turning speed and age. Despite these statistical controls, there may be a difference in turning speed for each walking task which was not evaluated.

The analyses reported in this study utilized turning speed during timed walking tests as the primary outcome and task, respectively. Future studies should consider exploring other metrics in timed walking to include cut-off scores for individuals at risk for functional decline. The overall distance score of these walk tests have established validity and reliability, however, quantification of turning metrics is not addressed. Establishment of normative values would allow evaluation of turning behavior with various number of turns and investigate a possible fatigue effect in turning.

5. Conclusion

Through this cross-sectional data analysis of 271 individuals with ages ranging from 20–88, turning speed proved to be a metric negatively affected by increasing age. Gait speed did not decrease with age until participants were in their 6th decade of life and gait speed remained similar in adults younger than 65 years of age. Considering turning speed is more responsive to age than gait speed, understanding how turning speed will change as people age is important in identifying who is at risk for decline in function and who has turning deficits that are not explained by age alone.

Highlights.

Turning speed normative values for U-turns were established per age decade.

Turning speed declined linearly with age, decreasing 0.66°/s each year.

Gait speed remained stable in younger adults and began declining at 65 years of age.