In Parkinson’s disease, dual-tasking reduces gait smoothness during the straight-walking and turning-while-walking phases of the Timed Up and Go test

Abstract

Background

Dual-task walking is used as a treatment, for gait assessment, and as an outcome measure in Parkinson’s disease (PD). Movement smoothness, i.e. its non-intermittency, is a movement cardinal feature. This study assesses whether dual-tasking reduces gait smoothness in PD alongside reducing speed, one of its well-known effects.

Methods

The Timed Up and Go (TUG) test, instrumented with an inertial sensor fixed to the back, was administered to 33 PD patients (15 females; age: median = 71 years; IQR = 10) to assess two walking types: straight-walking and turning-while-walking. The TUG test was completed in single-task and two dual-task modalities: cognitive (doing successive subtractions) and motor (holding a water glass). The angular speed spectral arc length metric (SPARC) and the Ln-DimensionLess Jerk (LDLJ), two smoothness measures quantifying the peaks and dips in the speed profile, were measured, along with the trunk angular velocity and the foot strikes number. ANOVA was used for hypothesis testing and estimated marginal means for post-hoc tests and effect sizes (ES).

Results

In straight-walking and turning, cognitive and motor dual tasks decreased gait speed (ES range = [0.476, 1.379]; p < 0.01) and increased the step number (ES = [0.402, 0.927]; p < 0.05). SPARC (ES = [0.221, 0.493]; p < 0.05) and angular LDLJ (ES = [0.451, 0.929]; p < 0.01) were lower in the two dual-task conditions in both phases, indicating reduced gait smoothness than in single-task. This worsening of gait smoothness was partially confirmed after ruling out the dual-task effect on speed and step number. In particular, anterior-posterior SPARC during turning was still low in cognitive (ES = 0.351; p < 0.01) and motor (ES = 0.283; p < 0.05) dual tasks.

Conclusions

In PD, dual-tasking decreases gait speed and increases the step number when walking straight and turning while walking. Independently of these effects, dual-tasking also reduces gait smoothness. As an independent feature of movement, when dual-task walking is the outcome measure, improving smoothness may represent a novel treatment aim in PD. As long as it is instrumented with an inertial sensor, the TUG test is valuable for studying different walking types.

Trial registration

NCT05904171 (ClinicalTrials.gov; date registration: 2023-06-06).

Background

Poor balance during walking causes disability and increases the risk of falling [1, 2], and patients with Parkinson’s disease (PD) commonly experience poor balance [3].

Dual-task, which consists of the simultaneous execution of two activities [4], such as walking while talking, is often used in rehabilitation as a therapeutic ingredient to improve balance [5] and reduce the risk of falling [6]. Balance and gait training with dual-task are frequently administered to PD patients, with meta-analyses showing that this training may improve balance in this condition [7, 8].

Dual-tasking is used not only for therapeutic purposes but also for assessment. Different mobility tests are completed in a dual-task modality for a more sensitive balance assessment (e.g [9]). and are used as outcome measures in clinical trials (e.g [10, 11]).

The Timed Up and Go (TUG) test is one of the most used tools for fall risk estimation (e.g [12, 13]). This test requires completing a circuit entailing fundamental motor functions, including walking in two primary ways: walking straight and turning while walking [14].

Alongside its single-task variant, the TUG test is also typically administered in a dual-task modality by adding a motor (e.g. carrying a full cup of water) or cognitive request (e.g. counting backwards) [15]. Dual-task is known to reduce walking speed and increase the TUG test duration (TTD). Surprisingly, only the total duration is usually considered when administering the TUG test with dual-task [15, 16]. Therefore, only a shorter TTD during dual-tasking indicates the patient’s improvement when using this test as an outcome measure.

Recording the TUG test, even with a single inertial measurement unit fastened to the trunk [17], allows splitting the TUG test into its different phases, and several mobility measures can be obtained for each phase [13]. Notably, these new mobility measures could work better than TTD in different scenarios [12, 18].

Accelerometers have been used since the sixties to study gait unintrusively and affordably, e.g [19]. , and in recent years, inertial sensors have increased in popularity. To date, inertial measurement units are also standard in a clinical setting to the point that they are recommended for reliable gait analysis [20].

Among the novel measures that can be efficiently gained from a trunk inertial measurement unit, particularly gyroscopes, are gait smoothness metrics [21, 22].

A movement with reduced smoothness is said to be jerky: it proceeds with a series of sharp, quick motions, continuously accelerating and decelerating until it stops and starts to flow again.

Smoothness is a third cardinal feature of movement, alongside amplitude and speed. More precisely, sound movement is on target (i.e. has the proper amplitude), has a proper speed (i.e. can be fast or slow as needed) and is smooth (i.e. processes in a continuous, uninterrupted way). Conversely, poor smoothness is an independent motor disturbance, unrelated to the reduced amplitude and speed (e.g [23])., distinctive features of the diseased movement (e.g [24, 25]).

Further evidence of smoothness as a primary, independent movement attribute comes from the clinic in which, for a long time now, the “irregularity of movement” observed in ataxias is recognised as separate from the “irregularity of movement” caused by paralysis [26, 27].

Given the importance of movement smoothness from a motor control and clinical perspective, the widespread use of dual-tasking, even for gait and balance assessment and the lack, to our knowledge, of studies investigating how smoothness is affected by dual-tasking, this study aims to assess if the dual task reduces gait smoothness in PD alongside (and net of) reducing gait speed.

Gait measures of walking trajectory (i.e. movement amplitude), speed and smoothness were extracted from the straight-walking and walking-while-turning phases of the TUG test and the dual-task TUG test, instrumented with a trunk inertial sensor.

Showing that dual tasks cause a smoothness impairment would be meaningful in research and in a clinical setting. Regarding the latter, showing that dual-tasking causes a genuine gait smoothness impairment offers new metrics for gait assessment and novel treatment goals.

Materials and methods

Participants

The data here analysed comes from an ongoing study aimed at assessing the cognitive, emotional and social effects of Argentine Tango in PD. The primary study intends to gather a sample of 24 patients who have completed multiple assessment sessions before and after the dance intervention. For the current work, the first 33 consecutive PD patients (see Table 1 for details regarding age, gender and years of diagnosis) were recruited. More precisely, only data from the baseline session were analysed, and patients were included here even if they dropped from the main study afterwards. The current one is thus an observational, cross-sectional study nested in a larger project (details on the primary project can be found on ClinicalTrials.gov: NCT05904171).

Table 1 Participants clinical characteristics

The study, which followed the Declaration of Helsinki on human studies [28], received approval from the local ethical committee (2023_02_21_02; P4419), and all participants gave their written informed consent to participate.

According to the main study inclusion and exclusion criteria, patients were included here, too, if:

• Older than 40 years.

• Diagnosed with PD according to clinical diagnostic criteria [29],

• Affected by mild to moderate Parkinson’s disease, i.e. scoring 1 to 3 on the modified Hoehn and Yahr Scale [30].

• Never practised Argentine Tango,

• Received no physiotherapy in the month before the study’s enrollment.

Patients were excluded per the following criteria:

• An additional major neurological disease (e.g. hemiparesis due to a stroke, traumatic brain injury),

• Dementia,

• A cardiorespiratory condition (e.g. heart failure) for which endurance activity is recommended under strict supervision,

• A psychiatric condition antecedent to Parkinson’s disease,

• The presence of deep brain stimulation or infusion pump for drug administration,

• Refusal to give consent to participate in the study.

Recruitment and testing were conducted at the IRCCS Istituto Auxologico Italiano – Ospedale San Luca, Capitanio in Milan (Italy).

Supplementary Materials 1 reports the power analysis run for the current study, given gait smoothness estimates for PD from a previous study [21].

Clinical, motor and neuropsychological assessment

The patients’ assessment included the following motor and neuropsychological tests collected by a neuropsychologist (LD, GS) and a physiotherapist (MA):

• TUG test instrumented with a trunk inertial measurement unit [13] (see below for details),

• 10 m walking test (10MWT) [31],

• Hoehn and Yahr rating scale [30],

• Unified Parkinson’s Disease Rating Scale (UPRDS) III [32],

• Frontal Assessment Battery (FAB) [33].

All patients were tested in the ON phase, diagnosed on a clinical basis, as a state in which patients responded well to antiparkinson medications [32]. In addition, before starting the assessment, they were asked if they felt themselves in the ON or OFF phase or if an OFF phase was approaching.

Participants completed the 3 m TUG as a “single task”, namely with no concurrent cognitive or motor task, and in dual-task modality, with a concurrent motor or cognitive task. In motor dual-task, participants were asked to complete the TUG test while carrying a plastic cup full of water. They were asked to make backward serial subtractions in the cognitive dual-task modality.

Each of the three TUG test versions (i.e. single task, cognitive and motor dual-task) was repeated three times, so nine TUG test repetitions were collected in a single experimental session.

Each session started with three repetitions of the single-task TUG test, followed by the cognitive or motor dual task. The order between cognitive and motor dual-task was randomised (coin tossing).

Single task, cognitive and motor dual-task TUG test was recorded with a commercial inertial measurement unit for clinical gait analysis (G-SENSOR 2 – motion analysis system, BTS SpA, Italy; see Supplementary Materials 1 for details) secured with an elastic band to the dorsal aspect of the trunk at the lumbar spine height. Only the signals from the triaxial accelerometer and gyroscope from the inertial sensor have been considered here (sampling frequency: 100 Hz; see below).

The 10MWT was administered in addition to the TUG test to calculate the walk ratio alongside gait speed and step frequency [34]. The walk ratio, the step-length/step-cadence ratio, brings out an actual law of motor control, i.e. that gait speed is increased by linearly increasing the step cadence and length. A walk ratio reduction is particularly expected in Parkinson’s disease, where step length is reduced and cadence is increased.

The Hoehn and Yahr rating scale and the UPRDS III were used to quantify the severity of parkinsonism. The Hoehn and Yahr scale is a single-item scale arranged into five categories from 1 (unilateral involvement, minimal impairment) to 5 (wheelchair-bound) [30].

The UPDRS III is the third domain of the UPRDS and addresses the motor severity of parkinsonism. It is made of 18 items, each scored from 0 to 4 (total score range: 0 to 132; the higher the score, the higher the Parkinsonism severity). Patients’ parkinsonism severity was classified into mild (≤ 32), moderate (> 32 and ≤ 58) and severe (> 58) per their UPDRS III total score [35].

Finally, the FAB was used to assess the presence of fronto-executive disfunctioning [36, 37], an early cognitive impairment in Parkinson’s disease [38, 39]. The FAB comprises six subtests that assess different frontal lobe functions, such as flexibility, motor programming, and inhibitory control. Each item is scored from 0 to 3, with the total score thus ranging from 0 to 18 (the lower the score, the more severe the dysexecutive syndrome).

Several cognitive tests have been used in Parkinson’s disease. For example, the Montreal Cognitive Assessment (MoCA) [40], likely one of the most used screening tools for cognitive impairment, has been used to assess the overall cognitive function in this patient population, e.g [41].

We have chosen the FAB rather than broader tests such as the MoCA because of the specific link between executive functions and walking. For example, impaired executive functions flagged by a low FAB total score are associated with freezing of gait in Parkinson’s disease [42]. Such a linkage is expected to be less apparent if a questionnaire of global cognitive functioning were used.

Signal analysis

Mediolateral and vertical angular velocity from the triaxial gyroscope and vertical and anterior-posterior acceleration from the triaxial accelerometer were used to split the TUG test into the following seven phases: (1) sit to stand, (2) standing to walking, (3) straight walking, (4) walking while turning, (5) walking back, (6) turning to sit and (7) sitting.

Since we are interested in assessing gait smoothness, only phases 3 (from now on shortened into “walking”) and 4 (from now on referred to as “turning”) are considered in the current work. The TUG test was used for assessing gait instead of a conventional walking test (e.g. 10MWT) precisely because with this test, as long as it is instrumented with an inertial sensor, it is possible to assess “at a stroke” two fundamental types of walking: straight walking and turning while walking.

The procedure for segmenting the TUG test into its seven phases is explained in Supplementary Materials 1.

Angular velocity from the gyroscope was low pass filtered with a zero-lag 4th -order Butterworth filter with a cutoff frequency of 20 Hz [43]. Acceleration signals were filtered according to [44] with a zero-lag, 4th -order Butterworth filter with bandpass 0.15 to 5 Hz.

In addition, angular acceleration was obtained by differentiating the angular velocity from the gyroscopes for time. Then, angular acceleration was filtered using the same bandpass filter of the linear acceleration.

The following measures were extracted from the TUG test:

• 1. TUG test duration (TTD, s),

• 2. Walking duration (s), which is the duration of the walking phase,

• 3. Turning duration (s), the duration of the turning phase,

• 4. Number of foot strikes in the walking phase,

• 5. Number of foot strikes in the turning phase,

• 6. Vertical angle (°) during turning.

Moreover, for the two phases and the three axes were also measured:

• 7. Root mean square (RMS) of the trunk angular speed (°/s),

• 8. Spectral arc length measure (named “SPARC”) from the gyroscope angular velocity,

• 9. Ln-dimensionless jerk (LDLJ) from the trunk linear acceleration (LDLJ_linear),

• 10. LDLJ from the trunk angular acceleration (LDLJ_angular).

The vertical angle during turning was calculated by mathematical integration of the vertical angular velocity.

SPARC, LDLJ_linear and LDLJ_angular, all dimensionless, are the metrics used here to quantify gait smoothness. Several smoothness measures have been developed, but SPARC and LDLJ are considered the most valid [45].

Many excellent studies detailing the mathematics behind SPARC and LDLJ, such as [45, 46], are available. Very briefly, SPARC and LDLJ build on the finding that a movement that briskly changes its trajectory many times is perceived as unsmooth (see Fig. 7 in [26] for a historical comparison of a smooth and an unsmooth movement).

Since movement changes are at stake when dealing with movement smoothness, looking for smoothness measures in the movement derivatives (such as speed) is obvious. In this regard, it has been known for a long time that the irregularities of an unsmooth movement are effectively brought into the foreground as “peaks and dips” in the speed profile (see, for example, Fig. 8 in [26]).

LDLJ is obtained from the jerk, i.e. the speed second derivative. The area under the curve of the squared jerk is calculated and normalised given the movement duration and maximum speed or acceleration to have a dimensionless jerk. This quantity’s logarithm (here, natural, ln) is computed to have LDLJ.

To have an intuitive idea of how jerk reflects the speed irregularities, note that the jerk’s integral, (i.e., the area) returns an acceleration, which is the velocity change in time, namely the peaks and dips in the speed time course.

SPARC is the arc length of the normalised magnitude spectrum of the movement speed. Intuitively, brisk peaks and dips in speed lead to high frequencies in the spectrum, lengthening its contour in this way.

Strictly speaking, the SPARC and LDLJ are measures of unsmoothness and decomposition of movement (the higher the SPARC and LDLJ numeral, the more ataxic the movement). However, in the smoothness literature, it is customary to speak of smoothness, rather than unsmoothness, measures, and thus SPARC and LDLJ are given as negative numbers so that the higher the number (i.e., the closer to zero), the smoother the movement.

SPARC and LDLJ are quite abstract indices, particularly for a clinical readership. To make these more friendly, Fig. 1 shows traces of two representative repetitions of the TUG test turning from two participants. The patient in Fig. 1A had poor anterior-posterior trunk smoothness during turning, while that in Fig. 1B had high smoothness.

Fig. 1

Low and high gait smoothness from the Timed Up and Go test: representative examples. Trunk anterior-posterior speed (upper panel) during the turning-while-walking phase of the Timed Up and Go test is shown for two patients, one with low (A) and another with high (B) smoothness of gait. Data are from two representative test repetitions matched for trunk speed root mean square (A: 35.1 °/s; B: 35.0 °/s), turning duration (A: 2.2 s; B: 2.7 s) and foot strike number (four for both). Speed offset was removed for this figure for graphical reasons. Middle panel: magnitude spectrum. Arrows mark the upper frequency for SPARC calculation. The horizontal continuous line is set at 0.05. Lower panel: squared anterior-posterior angular jerk (LDLJ_angular). At visual inspection of the speed traces (upper panel), it is apparent that there are more peaks and dips in the anterior-posterior speed profile in A than in B. More precisely, A has brisker and larger speed changes leading to steep-sided, sharp-pointed peaks, like the one indicated by the arrowhead. These more pronounced speed changes lead to an increased arc length of the speed spectrum (i.e. lower SPARC) and a larger area under the curve of the squared jerk (i.e. lower LDLJ)

SPARC and LDLJ were calculated according to [22] using the Python code provided with the paper.

Statistics

Linear mixed effects models [47] were used for the statistical analysis.

Central to this study is investigating if gait smoothness changes with dual-task. To this aim, the following regression model (simplified notation) was tested:

$$\begin{aligned} Smoothness = &\, {\beta\:}_{0}+\\ & \,{\beta\:}_{1}Task+ \\ & \, {\beta\:}_{2}Axis+ \\ & \,{\beta\:}_{3}Task\times\:Axis+ \\ & \,random+ \\ & \,\varepsilon \end{aligned}$$

(1)

With $\:Smoothness$, the smoothness metric, i.e. SPARC, LDLJ_linear or LDLJ_angular,

• $\:\beta\:$, the fixed effects coefficients, i.e. the intercept ($\:{\beta\:}_{0}$) and the regression slopes ($\:{\beta\:}_{1}$, $\:{\beta\:}_{2}$ and $\:{\beta\:}_{3}$),

• $\:Task$, the task condition (single task, cognitive or motor dual-task),

• $\:Axis$, the gyroscope and the accelerometer axis of measurement (vertical, mediolateral and anterior-posterior),

• $\:Task\times\:Axis$, the interaction between task and axis,

• $\:\text{random}$, the random effects

• $\:\epsilon\:$, model’s residuals

The model’s random structure included intercepts (subjects) and slopes (task and axis) so that maximal models were tested, as recommended [48]. The random slope structure was simplified in the case of “near singularity” since this potentially indicates overfitting [48].

Additional predictors were added to model 1 to rule out that the effects of the task condition on the smoothness metric were due to confounding effects rather than reflecting a genuine change in gait smoothness.

A model with the same structure as model 1 was also used to test if gait speed, with the trunk angular velocity RMS as the response variable, changed with the task and the measurement axis.

To assess if the task modality changed the TTD, the duration of the walking and turning phases, and the number of steps, the right side of Eq. 1 was simplified by dropping the axis predictor.

Type III ANOVA with Satterthwaite’s method was calculated to assess the statistical significance of the fixed effects of the regression models.

Estimated marginal means [49] were used for post-hoc t-tests with type I error probability equal to 0.05 as customary. When a set of tests was performed (e.g. posthoc tests after ANOVA, correlation analyses), the type I error probability was corrected for multiplicity according to Holm [50] for the number of tests within the set.

Effect sizes of the differences between means are also calculated [51].

The homogeneity of variance and residual normality assumptions of the linear regression were visually verified. If these assumptions were violated, the response variable was transformed, more commonly, ln-transformed. However, data are reported untransformed in the main text for easy understandability.

Median and 1st to 3rd quartile were preferred for descriptive statistics. Spearman correlation was run for preliminary analyses.

Signal 8.00 Cambridge Electronic Design Ltd was used to split the TUG test. Microsoft Excel 2016, R 4.2.0 and Python 3.11.5 were used for data analysis, and all the statistical analyses were run in R. R, and Microsoft PowerPoint 2016 was used for figures.

Results

Nineteen participants were older than 70, and only two were younger than 50 (median age = 71 years; 1st to 3rd quartile = 67 to 77 years). In addition, most of them suffered mild (n = 15 patients) or moderate parkinsonism (n = 16 patients) as assessed with the UPDRS III total score (Table 1). Participants were balanced for gender (18 males and 15 females).

About the gait impairment, at the sample level, the median gait speed at the 10MWT was 1.24 m/s (1.11 to 1.36 m/s), but the walk ratio was reduced (median = 4.9 mm/step/min; 4.4 to 5.1 mm/step/min), indicating that such a high walking speed was achieved at the expense of an increased step frequency given the step length.

When dynamic balance is assessed with the TTD, 17 participants completed the TUG test in more than 10 s, indicating some balance impairment. Only two participants had a TTD > 15 s. None of them had a TTD > 20 s.

For the FAB total score, about half of the sample suffered no or negligible impairment of the executive functions (8 persons had an FAB score of 17/18 and 7 of 18/18).

None of the participants had freezing of gait during the physical examination, nor did freezing of gait appear during testing.

Spatiotemporal parameters of the walking and turning phases of the TUG test completed in single and dual task modality

Dual-task increased the TTD (F_2,32.5 = 64.5; p < 0.001), which was longer in both cognitive (ES = 1.974; p < 0.001) and motor (ES = 1.265; p < 0.001) dual tasks, than in single task (Table 2). TTD was also longer in cognitive than motor dual tasks (ES = 0.710; p = 0.010).

Table 2 Dual-task effects on the timed up and go test duration, turning trajectory and number of steps

Similar effects of dual-tasking were found in the duration of the walking (F_2,32.0 = 44.9; p < 0.001) and turning (F_2,32.7 = 22.4; p < 0.001) phases (Table 2), but in the latter case with no difference between cognitive and motor dual-task.

Dual tasking also affected the number of foot strikes. In both the walking (F_2,29.0 = 17.6; p < 0.001) and turning (F_2,2.3 = 8.7; p = 0.001) phases, there were more foot strikes in the cognitive and the motor dual-task than in the single-task, with no difference between these two dual-task conditions (Table 2).

Regarding the trajectory during turning, the angle along the vertical axis was comparable in the three task conditions (F_2,262.2 = 0.9; p = 0.421).

Changes in gait speed and gait smoothness induced by dual-tasking: TUG test walking phase

The RMS of the trunk speed differed with the dual-task condition (F_2,33.2 = 29.0; p < 0.001), the axis of measurement (F_2,33.0 = 16.0; p < 0.001) and their interaction (F_4,720.7 = 4.4; p = 0.002).

Vertical trunk speed was reduced in cognitive (ES = 0.568; p = 0.002) and motor (ES = 0.476; p = 0.007) dual-task compared to the single task (Fig. 2). Mediolateral trunk speed was also reduced in cognitive (ES = 1.116; p < 0.001) and motor (ES = 0.565; p = 0.001) dual-task compared to the single task, and lower in cognitive than motor dual-task (ES = 0.551; p = 0.004). Changes in anterior-posterior trunk speed with the task and the measurement axis were similar to those of the mediolateral speed (Supplementary Materials 2).

Fig. 2

Gait speed and smoothness metrics in single, cognitive and motor dual-task: walking phase of the Timed Up and Go test. Black point graphs: estimated marginal means and their 95% confidence intervals (CI) from the interaction of the task condition and the measurement axis (i.e. mean values for each axis and task combination). White point graphs: estimated marginal means and 95% CI for the three task conditions (i.e. mean values given the task and irrespectively of the axis of measurement); white point graphs are reported only in the case of ANOVA reporting a significant task factor and insignificant interaction between task and axis. Walking: walking phase of the Timed Up and Go (TUG) test. Speed RMS: root mean square (RMS) of the trunk angular speed; SPARC: spectral arc length measure calculated from the trunk angular velocity; LDLJ: ln-dimensionless jerk from the linear acceleration (LDLJ_linear) or the angular acceleration (LDLJ _angular); Vert, ML, AP: vertical, mediolateral and anterior-posterior measurement axis. Single: single-task modality; Cog: cognitive dual-task; Mot: motor dual-task. SPARC and LDLJ are dimensionless (1). The black horizontal dashed lines mark the measures from the single-task modality. *: cognitive or motor dual tasks significantly differ from the single-task modality. °: cognitive and motor dual tasks are significantly different from each other

SPARC

Regarding gait smoothness (Fig. 2), SPARC changed with the task (F_2,31.3 = 13.2; p < 0.001), while it was comparable in the three axes (F_2,32.9 = 0.6; p = 0.546), and no interaction between task and axis was found (F_4,716.3 = 0.7; p = 0.616). SPARC was higher (i.e. the gait was smoother) in the single task than in the motor (ES = 0.221, p = 0.013) and cognitive (ES = 0.493, p = 0.001) dual-task. A difference was also found between the two dual-task modalities, with poorer smoothness in cognitive than motor dual task (ES = 0.272; p = 0.013).

LDLJ_linear

LDLJ_linear differed in the three tasks (F_2,31.4 = 25.7; p < 0.001), and, contrary to SPARC, LDLJ_linear also changed with the recording axis (F_2,32.9 = 10.6; p < 0.001) and the interaction between task and axis (F_4,716.4 = 2.7; p = 0.032).

Post-hoc testing exploring the task by axis interaction showed that vertical (ES = 1.070; p < 0.001), anterior-posterior (ES = 0.690; p < 0.001), and mediolateral (ES = 0.769; p < 0.001) LDLJ_linear was lower (i.e., gait was less smooth) during cognitive dual-task than in single task. In addition, vertical LDLJ_linear (but not anterior-posterior and mediolateral LDLJ_linear) was lower in motor dual-task than in single task (ES = 0.670; p = 0.001).

LDLJ_angular

The task condition affected the LDLJ_angular (F_2,32.2 = 32.0; p < 0.001), while this metric was unaffected by the measurement axis and the task and axis interaction. Per LDLJ_angular, smoothness was poorer in cognitive (ES = 0.929; p < 0.001) and motor dual task (ES = 0.497; p < 0.001) than the single task. LDLJ_angular was also lower in cognitive than motor dual task (ES = 0.431; p = 0.004).

Changes in gait speed and gait smoothness induced by dual-tasking: TUG test turning phase

The trunk speed RMS during turning depended on the task condition (F_2,33.1 = 42.1; p < 0.001), the measuring axis (F_2,32.9 = 749.1; p < 0.001) and the interaction between these two (F_4,720.4 = 8.2; p < 0.001).

Trunk speed was higher in the single task than in motor (ES = 0.911; p < 0.001) and cognitive (ES = 1.139; p < 0.001) dual-tasks (Fig. 3). No difference was found in speed RMS between the two dual-task conditions (p = 0.170).

Fig. 3

Gait speed and smoothness metrics in single, cognitive and motor dual-task: turning phase of the Timed Up and Go test. Same abbreviations as Fig. 2

Regarding the difference between the three axes, the speed RMS was higher on the vertical than anterior-posterior (ES = 8.71; p < 0.001) and mediolateral (ES = 10.56; p < 0.001) axes. Trunk speed RMS on the anterior-posterior axis was also higher than that on the mediolateral (ES = 1.85; p < 0.001).

Results from post-hoc tests run to explore the task by axis interaction did not change this pattern of effects (Supplementary Materials 2).

SPARC

SPARC depended on the task (F_2,31.8 = 9.4; p < 0.001) and the axis of measurement (F_2,33.0 = 356.0; p < 0.001). This smoothness index was comparable in the two dual-task modalities (p = 0.261; Fig. 3) and lower both in cognitive (ES = 0.387; p = 0.002) and motor dual (ES = 0.275; p = 0.002) tasks, as compared to the single task modality.

LDLJ_linear

Gait smoothness during turning changed with the task (F_2,32.5 = 11.2; p < 0.001) and the axis (F_2,32.9 = 7.3; p = 0.002) also when measured with the LDLJ_linear. LDLJ_linear was lower in cognitive (ES = 0.495; p < 0.001) and motor (ES = 0.389; p < 0.001) dual tasks, as compared to the single task. No difference in LDLJ_linear was found between the two dual-task modalities (p = 0.375; Fig. 3).

LDLJ_angular

Regarding LDLJ_angular, this also changed with the task (F_2,32.8 = 14.8; p < 0.001) and the axis (F_2,32.9 = 13.0; p < 0.001). In a comparable way to SPARC and LDLJ_linear, LDLJ_angular was lower in cognitive (ES = 0.566; p < 0.001) and motor dual task (ES = 0.451; p < 0.001) than in single task with no difference between the two dual-task conditions (p = 0.312; Fig. 3).

Independence of the dual-task induced gait smoothness decrement from gait speed and step number

Up to this point, it is shown that dual-tasking reduces the smoothness of gait when walking straight and turning, but it also decreases trunk speed and increases the number of foot strikes.

A preliminary analysis (see Supplementary Materials 3 for details) showed a positive correlation between smoothness metrics and trunk speed: the higher the metric of smoothness, hence the more smooth the gait, the higher the speed (Spearman’s Rho > 0.494; p < 0.05) and a negative correlation between smoothness and the number of foot strikes showing that the more the foot strikes, the lower the smoothness (Spearman’s Rho < −0.358; p < 0.05).

Given such findings, it is paramount to investigate if the worsening of the smoothness metrics found during dual-task reflects a genuine reduction in gait smoothness or the reduced gait speed and increased number of steps. Regression models with trunk speed RMS and foot strike number as predictors have been used to disentangle this issue.

SPARC

Regarding the walking phase, SPARC was comparable in the three tasks when considering the dual-task-induced differences in speed and the number of foot strikes.

More specifically, regression models with task, axis, and their interaction, as well as the trunk speed RMS and the number of foot strikes, pointed out only the significance of the latter two.

On the contrary, when the trunk speed and the foot strikes were added to the regression model for the turning phase, the interaction between the task and the axis of measurement was significant (F_4,790.3 = 2.81; p = 0.025).

Post-hoc testing showed that SPARC was lower, i.e. gait smoothness was poorer, on the anterior-posterior axis in the cognitive dual task compared to the single task (ES = 0.351; p = 0.006). Anterior-posterior SPARC was also lower in the motor dual task than in the single task (ES = 0.283; p = 0.048).

LDLJ_linear

Differently from SPARC, when the RMS of the trunk speed and the number of foot strikes were included in the regression model, the task condition (F_2,36.1 = 4.3; p = 0.022), the measurement axis (F_2,35.1 = 11.5; p < 0.001) and the interaction between these two (F_4,717.6 = 3.9; p = 0.004) remained significant predictors of LDLJ_linear in the walking phase.

Post-hoc testing exploring the significant interaction showed that the difference in LDLJ_linear between the three tasks only holds for the vertical axis, with gait less smooth in the cognitive dual task than the single task (ES = 0.565; p < 0.001). Even if not significant, vertical gait smoothness was also lower in the motor dual task than in the single task condition (ES = 0.332; p = 0.050).

Finally, in the turning phase, the regression model with speed and foot strikes as covariates resulted in a significant interaction between task and axis (F_4,729.2 = 2.54; p = 0.039).

Post-hoc comparisons revealed that anterior-posterior LDLJ_linear was lower during cognitive than motor dual-task (ES = 0.369; p = 0.025). Anterior-posterior LDLJ_linear was also lower in the cognitive dual task than in the single task, but this difference was insignificant.

LDLJ_angular

In the walking phase, the task condition remained a significant predictor of LDLJ_angular (F_2,36.4 = 8.53; p < 0.001) even after including in the regression model the trunk speed (F_1,504.9 = 23.2; p < 0.001) and the foot strike number (F_1,341.1 = 542.3; p < 0.001).

Post-hoc testing confirmed LDLJ_angular was lower in cognitive dual-task (ES = 0.345; p < 0.001) than in single task. No significant difference was found between motor dual-task and single-task, even if LDLJ_angular was lower in the former. LDLJ_angular was lower in cognitive than motor dual-task (ES = 0.244; p = 0.033).

The task was also significant in turning (F_2,51.8 = 3.9; p = 0.026), along with the measurement axis (F_2,67.2 = 12.3; p < 0.001).

Once more, gait smoothness was poorer in cognitive dual-task than in the single task, although this difference was insignificant (ES = 0.216; p = 0.055). Smoothness was significantly worse in cognitive than motor dual tasks (ES = 0.173; p = 0.032). No difference was found between motor dual-task and single-task (p = 0.589).

The full results of the statistical analysis and the regression models are provided in Supplementary Materials 2.

Relationship between gait smoothness and executive functions in Parkinson’s disease

Dual-tasking reduces gait smoothness, even after ruling out the confounding effect of the dual-task-induced gait speed reduction and increased foot strikes. This effect seems more marked for cognitive than motor dual-task, a finding suggesting that cognitive resources could be needed to walk smoothly in Parkinson’s disease.

This hypothesis has been further evaluated by investigating the correlation between the clinical index of the executive functions, namely the FAB total score, and the smoothness metrics considering only those metrics that changed with the task condition (even when trunk speed and the foot strike number were included in the regression models).

A significant, positive correlation was found between the anterior-posterior SPARC during turning in the single task modality and the FAB total score (Spearman correlation: rho = 0.549; p = 0.002; Fig. 4), indicating that those patients presenting with a frontal-executive impairment also had poorer gait smoothness.

Fig. 4

Correlation between gait speed, gait smoothness and executive functions. Single, Cognitive, Motor: single task modality, cognitive and motor dual-task; AP: anterior-posterior; RMS: root mean square; SPARC: spectral arc length measure (dimensionless, 1); FAB: Frontal Assessment Battery. The FAB scores are split into < and ≥ the median (vertical dashed line). Rho: Spearman’s rho; p: p-value of the Spearman correlation. The boxplots show the speed and SPARC values in the participants’ sample split according to the median FAB value to make the association between walking metrics and executive functions more apparent

Importantly, this relationship was confirmed when a partial correlation was run, including the UPDRS III total score, to rule out the effect of the Parkinsonism severity (Spearman partial correlation: rho = 0.516; p = 0.012).

A comparable relationship, again given the parkinsonism severity, was found between anterior-posterior SPARC and the FAB total score in cognitive (rho = 0.460; p = 0.032) and motor dual-task (rho = 0.546; p = 0.007).

On the contrary, no significant relationship was found between the FAB total score and LDLJ_linear or LDLJ_angular.

There was no correlation between trunk speed and the FAB (Fig. 4) or the UPDRS III total score in any of the three axes and the two TUG test phases (see Supplementary Materials 4 for this correlation analysis).

Discussion

The present study shows that cognitive and motor dual-task consistently reduce gait speed, thus lengthening the duration of the straight-walking and turning-while-walking phases of the TUG test. Dual-task also increases the number of steps in walking and turning, and dual-task, cognitive and motor, generally reduces gait smoothness, i.e. increases movement decomposition during walking.

This smoothness reduction is confirmed, at least to some extent, net of the dual-task-induced changes in gait speed and number of steps.

After controlling for these two confounders, gait smoothness while turning was still reduced along the anterior-posterior axis in cognitive and motor dual-task, both when measured with SPARC and LDLJ, linear and angular.

Regarding the phase of straight walking, after ruling out the changes in speed and step number, no dual-task-induced change in smoothness was confirmed with SPARC. On the contrary, smoothness was poorer in dual-task when quantified with LDLJ.

Lastly, gait smoothness is only associated with clinical ratings of executive function impairment when assessed by SPARC and only for the turning phase.

Why study gait smoothness?

In clinical practice, it is well recognised that movement smoothness is an independent feature of movement, with ataxia, whose hallmark is the decomposition of movement and thus the smoothness reduction, being a clinical syndrome by itself [26, 27].

From an experimental point of view, in human movement science and motor rehabilitation, examining the smoothness of movement corresponds to examining what is called, in jargon, “motor control”, i.e. how the nervous system recruits and coordinates motor centres, motor units and muscles to carry out a movement.

The smoothness measures used here, i.e. SPARC and LDLJ, are grounded in a well-defined motor control theory. Movement is considered to be made up of submovements that are only scalable in size and duration [52]. In disease, submovements can be of inappropriate amplitude and duration and be recruited at the wrong time. Peaks and dips appear in the speed profile of the overall movement, i.e. movement smoothness is reduced, and the movement appears decomposed and jerky. Hence, reduced smoothness of movement (or, conversely, decomposition of movement) flags a motor coordination problem [53].

Having a high validity as motor control measures (i.e. pointing out that some have gone wrong in the programming of movement), gait smoothness metrics could add novel information for the diagnosis of gait impairment and for setting the gait impairment’s prognosis. In addition, they could be set as novel outcome measures for assessing gait treatment effectiveness.

When used as treatment outcomes, improved smoothness suggests better motor programming, eventually suggesting a deeper, “intrinsic” recovery has occurred [54].

In diagnostic terms, Parkinson’s disease patients with severely reduced smoothness could be those with a higher risk of falling or suffering more freezing episodes during the day, given the association between impaired smoothness and executive impairment (see next for a discussion about this finding) and the association between impaired executive functions and gait impairment.

In prognostic terms, patients with particularly low levels of gait smoothness in dual-task walking could respond less to motor learning (and hence to physiotherapy), again because of the link between poor smoothness and executive impairment.

However, it is worth stressing that applying gait smoothness metrics in assessment and setting the prognosis remains highly speculative, an actual hypothesis for future studies.

What does the gait smoothness deterioration induced by dual-tasking in Parkinson’s disease consist of from a clinical standpoint? It can be hypothesised that in this condition, dual-tasking induces (i) gait ataxia, (ii) dyskinesias, (iii) tremor or (iv) festination. At the moment, it is impossible to clarify further which of these four possibilities the reduced smoothness is an indicator of. The other way around, all these four motor disturbances would become apparent as reduced trunk smoothness.

Regarding gait ataxia, it is worth mentioning here the involvement of the cerebellum in Parkinson’s disease (e.g [55]). Tremor is a cardinal sign of parkinsonism, and chorea and dystonia (i.e. dyskinesias) are common motor complications of the disease [32]. What can be said in this regard is that during testing, no overt tremors or dyskinesias appeared during the dual task. Similarly, no overt festination occurred, a walking modality characterised by short and high-frequency steps, such as that occurring in a freezing of gait episode [56].

Why study dual-task in walking?

A reason for studying dual-tasking while walking comes from the consideration that walking while completing concurrent motor or cognitive tasks, such as talking, using a smartphone or carrying a glass of water, is a common requirement in daily life [4]. We can say that this is how we move: we do other things while walking instead of focusing on our gait.

In the clinic, dual-task walking is administered for assessment, as an outcome test to evaluate treatment effectiveness, and as an exercise in a gait training program.

When used for assessment, dual-tasking can unmask a walking impairment [57], which could go unnoticed. Gait speed and step length, which are notoriously reduced in Parkinson’s disease, are further reduced by dual-task [4, 58], and, in this condition, dual-task also reduces walking regularity [59] and disrupts gait symmetry [60].

Dual-task walking is also administered to assess the treatment’s effectiveness [10, 11]. A beneficial gait treatment would decrease the walking speed drop typical of dual-task walking. Since it is shown here that dual-tasking also causes a genuine reduction of gait smoothness, in line with what has been proposed in the previous section, gait smoothness metrics could be employed as novel outcome measures in single (see previous section) and dual-task walking.

Studying dual-task in walking is also interesting since it is often used as a treatment. In line with setting gait smoothness in dual-task walking as a treatment outcome, it can be speculated that therapists should train patients to keep their gait speed and gait smoothness high when exercising in dual-task. However, while the former objective is feasible, more research is needed to monitor gait smoothness in clinical practice (see the study’s limitations).

So, to sum up, if it is worth investigating gait smoothness from a motor control perspective, given the importance of dual-task in assessing and treating gait impairments, it is reasonably even more crucial to assess gait smoothness in dual-task walking.

Dual-tasking, Gait smoothness and executive functioning

A relationship between reduced gait smoothness (i.e. low SPARC) and impaired executive functions (i.e., low FAB scores) has been demonstrated here, which is independent of Parkinsonism severity. In addition, the reduction in gait speed and the worsening of smoothness induced by dual-task is particularly marked in cognitive dual-task.

Executive functions include, among others, abstract reasoning, attention, inhibitory control and motor programming [36, 61]. If, as already discussed, high movement smoothness indicates flawless motor programming, it is thus not surprising that smoothness is reduced in the case of executive impairment.

Regarding the more marked gait impairment in cognitive rather than motor dual-task walking, a certain level of attention is needed to complete the cognitive assignment of the TUG test in cognitive dual-task. It can be put forward that attention is diverted from walking to thinking. However, because of gait impairment, walking itself needs increased attention. If this is the case, reduced attention to walking makes overt a gait impairment.

Of course, a similar mechanism is at play in the TUG test with motor dual-task. Attention is diverted from walking to complete the glass transportation successfully.

Even if it is impossible to say if a secondary cognitive task, such as that used in the dual-task TUG test administered for the current study, is more attention-demanding than the motor one, it is noteworthy that previous pieces of evidence reported that this could be the case (e.g [62]).

Moreover, it can be supposed that, from an opposite point of view, completing the TUG test in a cognitive dual task modality is an actual prefrontal lobes task, much more than transporting a glass while walking.

It is very well known that patients with an executive impairment have more difficulties in unconventional situations in which they are asked to find for themselves novel solutions [36]. Making backward serial subtractions while walking is precisely a task with these features. On the contrary, simultaneously walking and transporting is not exceptional, being customary in everyday life. If this theory holds, it is clear that more attention is drained from walking in favour of the cognitive secondary task than the motor one.

Redirecting attention from walking to a secondary task is a key component of this model. In this respect, dual-tasking is frequently used for rehabilitation to divert attention from the lower limbs and the environment (such as uneven terrain).

The relationship shown in this study between gait smoothness, dual-tasking, and executive functions aligns well with previous reports, which also highlight a linkage between executive functions and walking. For example, disturbances in frontal cortical regions have explicitly been associated with walking impairments, such as freezing of gait, in parkinsonism [63].

Why administer the TUG test if interested in gait?

This deliberate choice has been made since the TUG test allows for assessing at once two gait modalities: straight walking and turning while walking.

However, it should be stressed that these forms of walking derived from the TUG test are clearly different from straight walking at a constant speed assessed with the 10MWT. Instead, the TUG test and the 10MWT complement each other in gait assessment.

Turning requires a better balance than straight walking, as suggested by a more robust association between a turning impairment and falls compared to an impairment of walking in a straight line [12]. In other words, turning can be more challenging regarding the ability to walk without falling (i.e. walking balance precisely) [3, 13].

Regarding straight walking from the TUG test, this can be considered walking between two real balance challenges (two balance “trips”, actually): the end of standing up and starting to turn.

Which smoothness metric? LDLJ, SPARC or both?

Several indices have been used to quantify smoothness. However, not all of these are considered valid smoothness measures, and over the years, a consensus has developed that SPARC and LDLJ should be preferred for smoothness quantification [64, 65].

Nevertheless, as indicated previously [23] and confirmed by this work, SPARC and LDLJ are not equivalent smoothness measures and should not be regarded as interchangeable.

Regarding the current study, the most marked changes induced by dual-task are those highlighted by LDLJ_angular, which shows between-task differences along all three measurement axes and in walking and turning, irrespective of speed and number of steps. Next, LDLJ_linear also shows differences between tasks in the two TUG test phases. SPARC shows a between-task difference only regarding the turning phase. So, in terms of responsiveness to dual-task, the three metrics can be ranked high to low: LDLJ_angular, LDLJ_linear and SPARC.

However, this last smoothness metric seems to have the highest construct validity, where the hypothesis about the construct is that movement smoothness, as an indicator of motor coordination and programming, depends on the functioning of the prefrontal lobes [66]. In fact, this is the only metric that shows the linkage between dysexecutive syndrome severity and gait smoothness reduction.

As a final technical note, it has been suggested that when inertial measurement units are used, the SPARC algorithm can be applied directly to the angular velocity returned by the gyroscope. In contrast, the orientation of the inertial sensor should be reconstructed before deriving the triaxial LDLJ from the acceleration signals [22].

Here, we deliberately did not implement the procedure for correcting the accelerometer orientation (e.g [18]). for simplicity and immediacy reasons and since this choice has already been put into practice in other studies (e.g [44]).

Limits and future developments of the study

A primary limitation of the study was that it did not record the effects of dual-tasking on concurrent tasks and, above all, cognitive tasks. Given previous literature, some participants possibly deteriorated their gait while performing well in cognitive terms, while others performed poorly on both assignments.

Regarding the dual-task effects, it should be noted that, generally, the effects induced by completing a movement in cognitive dual-task can be pretty complex. When there is the simultaneous performance of a cognitive and a motor task (what has been called here “cognitive dual-task”), cognitive-motor interference can lead to different outcomes, from the worsening of the motor task, which is what we have only considered here, to improving both the cognitive and motor task [67].

It should also be noted that no prioritisation order between the two tasks (e.g. counting backwards or walking) was specified.

Further investigation on the differences between SPARC and LDLJ as smoothness metrics seems necessary, and studies about the construct validity of these measurements should be encouraged. Of particular help in this regard could be comparing SPARC and LDLJ in patients known to suffer an ataxic disturbance, sensory or cerebellar in the first place to identify which of the two smoothness metrics is the one working better for quantifying ataxia.

The relationship between movement speed and smoothness is known from the literature [52] and confirmed by the data presented in this work.

Since dual-task decreased gait smoothness and gait speed, we used regression models to understand if dual-task affects smoothness net of its effects on speed, i.e. to understand if dual-task decreases smoothness by itself or if it decreases speed and, from this, the smoothness reduction.

However, from an opposite point of view, it has been suggested that movement speed increases, given that movement becomes smoother [52]. This latter consideration, i.e. that high smoothness causes high speed could even lead to believing that smoothness should be studied as it is, irrespective of the associated changes in speed.

Nevertheless, in clinical practice, it is well known that persons suffering from ataxia can move with good speed, indicating that the scenario artificially created with regression can be naturally found and should be the one to pursue.

Developing smoothness metrics independent of speed would be the next step from a clinical standpoint. This novel line of research would be precious given that when a patient with a motor disturbance gets better, they often improve speed in the first place [18, 68].

The effects of dual-task on gait smoothness in Parkinson’s disease could also be assessed with an optoelectronic system to understand to what extent LDLJ calculation on the raw acceleration signal can be tolerated and to reach a fuller understanding of gait impairment flagged by the reduced smoothness (e.g. gait ataxia vs. tremor vs. dyskinesias).

Finally, we considered here a single test of cognition, i.e. the FAB, to assess the relationship between cognitive functioning and walking. This choice was taken for an apparent reason. We were interested in studying the relationship between executive functions in Parkinson’s disease and gait smoothness, and the FAB is a criterion standard in this regard.

Nevertheless, a further development of this research would be studying in greater detail the association between cognitive functions and gait smoothness in dual-task walking and gait smoothness in general.

While we would avoid investigating the relationship of gait metrics with tests of overall cognition like the MoCA, as already argued in the Methods, the relationship with the executive functions can be explored in greater detail with more dedicated tests [61].

Conclusions

Walking in a cognitive and motor dual-task modality reduces the gait speed and increases the number of steps. Moreover, exclusive of these effects, gait smoothness is reduced in Parkinson’s disease.

The two primary smoothness metrics, LDLJ and SPARC, are not wholly interchangeable when studying gait smoothness, and further research is needed to understand the different aspects of gait smoothness that are preferentially quantified from the former or the latter.

When quantified with SPARC, poor gait smoothness is associated with impaired executive functions, an expression of the link between the prefrontal lobes, motor programming, and walking.

What the smoothness worsening induced by dual-task is in Parkinson’s disease from a clinical point of view remains to be understood. So far, it cannot be disentangled if this is an expression of gait ataxia, tremor or dyskinesia.

The TUG test, as long as recorded with an inertial measurement unit secured to the trunk, is invaluable for assessing two primary types of walking: straight walking and turning while walking.