Introduction

The multi-system pathology of Parkinson’s disease (PD) is well recognised. Evidence supports the independence of certain gait characteristics from dopamine, suggesting that gait also has a complex pathophysiology in PD, with spatial parameters responding more favourably to dopamine than temporal parameters [1, 2].

Recent research has focussed on the contribution of gait variability as a mechanism for understanding healthy and pathological gait. Gait variability, typically expressed as stride-to-stride fluctuations in gait, reflects the underlying neural control of locomotion with the ability to generate rhythmical stepping movements without conscious control [3]. Stride time variability is commonly reported and thought to relate to rhythmicity. It is sensitive to measures of executive function in healthy older adults, in adults with pathology and when walking under dual task conditions [4, 5]. Double limb support (DLS, time spent with both feet in contact with the floor) is an alternative measure of variability that reflects postural control mechanisms contributing to the dynamic act of walking [3], and increases in healthy, older adults, and adults with cerebellar dysfunction and basal ganglia disease [5]. The characteristic of variability is not limited to gait, and is commonly used as a parameter of general bilateral co-ordination. Camicoli [6] found increased coefficient of variation (CV) in time for simple and choice reaction tasks, and Almeida revealed timing deficits during the ‘anti-phase’ movement of a bimanual co-ordination task compared with controls [7].

Stride time variability is present in de novo PD and has been shown to be independent of bradykinesia and gait speed, and therefore of putative dopaminergic pathways [8]. Other neurochemical pathways may contribute to gait variability in PD [9], making it a potentially useful marker of non-dopaminergic pathology.

Previously we identified that personal, motor, cognitive and affective characteristics contribute to walking speed and gait interference in PD, highlighting the multifactorial nature of gait control, which in turn is influenced by task demand. Earlier reports demonstrate the importance of depression, fatigue, executive function and fear of falling to performance under single and dual task conditions [10–12]. We wished to extend this work to explore the explanatory characteristics of gait variability in PD, using two distinct parameters to represent the role of temporal control through generation of the locomotor pattern (stride time) and postural control (DLS time). Participants were evaluated on and off medication to explore the contribution of dopamine to performance. We also examined performance under single and dual tasks, because we anticipated that the explanatory characteristics for variability may differ under complex conditions when gait is forced to be controlled more automatically, reducing the ability to compensate, thus revealing impact of pathology. Our primary hypothesis was that some of these characteristics would reflect non-dopaminergic pathways.

Methods

Participants

A convenience sample of 50 people with idiopathic PD was recruited from a local movement disorders clinic. Inclusion criteria were clinical diagnosis of idiopathic PD, disease severity between I and IV on the Hoehn and Yahr scale [13], treatment with levodopa therapy, self-report of a definite response to medication (without prolonged off periods), absence of any other neurological problem or any severe co-morbidity likely to affect gait, absence of dementia (defined by DSM-IV criteria and with a score above 24 on Mini Mental State Examination (MMSE) [14]), adequate sight and hearing with glasses or hearing aid if required, independently mobile indoors without a walking aid, no severe dyskinesias (above 2 on Modified Dyskinesia Scale) [15] and age 80 years or less. All participants gave informed written consent. Ethical consent for the study was granted by Newcastle and North Tyneside Local Research Ethics Committee, UK.

Explanatory characteristics

Explanatory characteristics were selected based on previous research [11], representing the following domains:

Personal Age, gender, years of disease duration and total daily intake of dopamine [16].

Motor Motor disease severity was measured with the Unified Parkinson’s Disease Rating Scale section III (motor subscale) (UPDRS III) [17] and freezing of gait with the New Freezing of Gait Questionnaire (NFOG-Q) [18].

Cognitive The Hayling and Brixton was used to measure cognitive flexibility [19]. Attention was measured with two domains of the Test of Everyday Attention (TEA): telephone search (time per target), which measures visual selective attention, and telephone search whilst counting (dual task decrement), which measures divided attention [20]. Different versions of each TEA subtest were used for testing on two occasions.

Affective Fatigue was measured using the Multidimensional Fatigue Inventory (MFI) General Fatigue Subscale, a 20-item questionnaire that evaluates physical and mental fatigue symptoms [21], and depression was assessed with the Hospital Anxiety and Depression Scale (HADS) [22].

Experimental protocol

Participants were tested at home on two occasions. On-medication testing took place around 1 h after medication intake, and off-medication testing took place before the first daily dose of medication. On both occasions, testing took place at a similar time of day. Participants were randomised using opaque, numbered envelopes to receive their first assessment either on or off medication and completed a functional walking task which has been described elsewhere [11]. Briefly, the task involved walking 6 m towards a bench (single task), collecting a tray with two cups of water placed on it filled to a standardised level, turning 180° and walking back to the start position carrying the tray and cups (dual task). Three walking trials were performed. Participants were instructed not to prioritise either the tray carrying or the walking task but rather to concentrate on the task as a whole. Participants repeated the walk if festination or freezing was observed. Gait parameters were collected using a stride analyzer (B & L Engineering, Tustin, CA, USA) which samples footswitch recordings at a rate of 500 Hz over the central 4 m of the walk to avoid acceleration and deceleration. Gait speed, stride time and double limb support (DLS) time were recorded.

Data analysis

Gait variability for stride time and DLS time was calculated using the coefficient of variation (CV) as CV = (SD/mean) × 100, for each stride [4]. Left and right footswitch recordings were pooled for all measures of variability to increase the number of data points used. An average of 4.5 strides per trial were collected, and these were pooled across three trials and used to calculate the CV. Mean values for gait characteristics were calculated from three walking trials. Univariate analyses were used to describe the data. Student’s t test was used to compare baseline measures on and off medication. Pearson’s product correlation coefficients were calculated to explore associations between independent characteristics and gait variability. Hierarchical multiple regression was used to identify predictors of stride time variability and DLS time variability under single and dual task conditions, both on and off medication. Nine variables were selected on the basis of previous research and bivariate analysis. The first block of variables included four characteristics known to influence gait performance that we wished to control for: age, fatigue (MFI), depression (HADS-D) and gait speed. The second block included five variables that we hypothesised would influence gait variability: motor severity (UPDRS III), executive function (Hayling and Brixton), selective attention (TEA telephone search) and divided attention (TEA dual task decrement), which were expected to change on and off medication.

Residuals and collinearity diagnostics were examined and assumptions for regression analysis met for all data (including CV). Standardised beta coefficients, part, and part squared correlations were used to evaluate the relative contribution of each predictor to the variance of dependent measures. The sample size was pragmatic, however we aimed to recruit a minimum of five participants for each independent variable entered into regression analysis. The alpha level was set at 0.05. SPSS (version 17) for Windows was used to analyse the data.

Results

Subject characteristics

Fifty people with PD with moderate disease severity participated in the study. Median Hoehn and Yahr score on medication was 3, indicating moderate disease severity (2 people scored 2, 12 scored 2.5, 32 scored 3 and 4 scored 4). Forty participants (80%) experienced episodes of freezing, and 27 (54%) reported at least one fall in the previous 6 months. HADS-D scores were higher than 7 for 34 people (68%), indicating depression. All participants were taking levodopa therapy; 21 participants were also taking a dopamine agonist and 12 participants a catechol-O-methyl transferase (COMT) inhibitor (Table 1).

Table 1 Demographic characteristics and subject characteristics on and off medication
Full size table

Effects of dopamine on performance

Motor and cognitive performance was worse off medication under both single and dual task conditions. UPDRS III scores significantly increased, indicating loss of motor control, and Hayling and Brixton scores deteriorated, denoting poorer cognitive performance (p < 0.05). Scores for selective attention (TEA telephone search) worsened but did not reach significance, whilst scores for divided attention (TEA telephone search whilst counting) significantly deteriorated. Gait speed significantly decreased off medication for both single and dual task conditions. Stride time variability increased by 27.6% for single task and 19.2% for dual task, and DLS time variability increased by 10.2% off medication for single task, and by 2.0% for dual task, which was not significant (Tables 1, 2).

Table 2 Gait parameters on and off medication
Full size table

Stride time variability

The first block of variables (age, gait speed, depression and fatigue) did not significantly predict stride time variability under single task conditions. Inclusion of the second block of variables (motor severity, executive function and attention) explained 32.6% of variance on medication and 58.8% of variance off medication. Motor severity emerged as the only significant predictor in both models. The results were comparable for dual task conditions, with the exception that depression emerged in the model as a significant predictor along with motor severity both on and off medication, with both factors explaining 38% of variance on medication and 58.1% of variance off medication. Depression made a unique contribution of 9.0% on medication and 5.0% off medication (Table 3).

Table 3 Summary of regression analyses predicting stride time variability and double limb support time variability in single and dual task conditions, on and off medication
Full size table

DLS time variability

As with stride time variability, block 2 (motor severity, executive function and attention) accounted for a greater portion of variance. For single task conditions, motor severity, executive function and attention explained 66.4% of variance on medication and 49.9% of variance off medication. Motor severity again emerged as the only significant predictor in both models. A different picture emerged for dual task conditions. Lower scaled scores for visual attention, younger age and motor severity emerged as significant predictors, with the final model accounting for 84.7% of the variance. Visual attention scores made a unique contribution of 3%, and age a unique contribution of 2% to the model. For dual task conditions off medication, motor severity was the unique contributor, with the final model accounting for 74.0% of the variance (Table 3).

Impact of freezing on outcome

To discern the effect of gait freezing on prediction models, we ran the regression analysis on the subset of freezers (n = 40). The results were highly comparable, and the analyses are not reported here.

Discussion

This study is the first, to our knowledge, to identify the explanatory characteristics of gait variability in PD by considering a broad range of personal, motor, cognitive and affective symptoms identified in previous studies to contribute to gait speed. We hypothesised that at least some characteristics that influence variability would be independent of dopamine, which was supported by our results. The explanatory characteristics of stride time and DLS were the same on and off medication during the single task, with motor severity (UPDRS III scores) accounting for most of the variability. However, when the system is sufficiently stressed under dual task conditions other predictors emerge, explaining a greater portion of the total variance. Dual task performance is challenging for people with PD, who compensate for motor dysfunction by relying on attention and cognitive strategies to maintain performance, something which becomes difficult during dual task performance [4].

The effect of dopamine on gait variability was selective. Stride time variability significantly improved on medication [23], but DLS variability did not [24], thereby implicating different neurochemical substrates in the control of disturbed postural control of gait. The mechanisms underlying the non-dopa effects seen for DLS are speculative but may include mediation via cholinergic pathways. The independence of DLS CV from dopamine and its relationship with dynamic postural control during gait is consistent with a role of the pedunculopontine nucleus (PPN) in balance control in humans [25].

As anticipated, during single task conditions (with the potential for gait to be under cognitive control), motor severity contributed significantly to both stride time and DLS variability, irrespective of dopamine status. However, the relative contribution differed for each characteristic, with a smaller contribution on medication than off medication for stride time variability, and the opposite for DLS variability. This points to the complex pathophysiology of gait in PD and the compromised effect of dopamine on gait performance.

Under dual task conditions, when gait was forced to become more automatic and with limited attentional reserve to call on, other characteristics emerge alongside motor severity and become more relevant to maintain control. For stride time variability, over half of the total variance under dual task conditions was explained by motor severity and depression, both on and off medication. The role of depression in PD gait supports earlier findings [12] and suggests that different neurochemical substrates may be involved [9, 26]. Depression is common in PD, with frequency of around 45% [27]. Although the participants presented with HADS-D scores just above the normal range (0–7), 52% were depressed according to HADS-D criteria [22]. A tentative link, potentially relevant to this study, is the role of norepinephrine in freezing of gait [9], supported by animal models that link norepinephrine deficiency in the locus coeruleus to this phenomenon [28]. The influence of depression on stride time variability may have masked the influence of cognitive function and executive function, which we expected to see especially under dual task conditions in PD off medication. Overall, these results support further examination of depression as an explanatory characteristic for gait variability in PD.

For DLS variability, a different picture emerged, which is not surprising in view of its distinct contribution to locomotion. Both attention and age contributed to DLS, although when on medication rather than off medication, which we had not anticipated. It is possible that attention, enhanced by dopamine [29], became more accessible when on medication and was able to be utilised, whereas when off medication, motor severity dominated. The influence of younger age on increased gait variability is difficult to interpret but may reflect an interaction with attention, motor severity or dopamine response, which was not explored in the analysis. Importantly, motor severity, age and selective attention explained 84.7% of the variability in DLS on medication, whereas motor severity was the only significant predictor for DLS off medication, explaining 47% of variability. In addition, other characteristics not included could contribute to DLS variability off medication, such as visual field acuity. The complex interaction of neurochemical contributors on and off medication is reflected in these findings. More work to untangle these complexities is needed if we are to understand how best to manage these distressing aspects of motor dysfunction, and optimise safety.

In this study, UPDRS III was used as a proxy measure for disease severity. Not all scale items are dopa responsive, as evidenced by the retention of UPDRS III as an explanatory variable on medication for both stride time and DLS variability. This suggests a role for other potential contributors which are non-dopa. In addition, the increased role of the UPDRS under dual task highlights the inability to compensate for basal ganglia dysfunction with cognitive control.

This study confirms previous reports of the influence of dopaminergic medication on mean values for PD gait parameters, demonstrating a preferential effect for improvement of gait rhythmicity rather than postural control [1, 2]. Positive effects of dopaminergic medication on cognitive function and dual task attention were also observed, in concordance with earlier work supporting a beneficial but selective l-dopa response [29]. Regression results demonstrated the independence of gait speed from gait variability, thereby providing distinct but complementary information.

There are some important limitations that should be addressed. Participants were tested in their own home to enhance the ecological validity of the findings, which meant a short distance over which variability data could be collected, although variability outcomes are comparable to earlier research. Calculation of variability from stride rather than step data did not allow for the effects of gait asymmetry, although pooled data is commonly reported in variability studies. The study may have been underpowered to pick up significant findings in regression analysis. In view of the dominance of PD freezers in this study, findings are not generalisable to non-freezers. Finally, we used a practical definition to describe off-medication status, taking into consideration the longer half-life of dopa agonists, which some participants were taking. Nevertheless, on- and off-medication results showed significant differences.

Overall, this study provides support for a significant contribution of non-dopaminergic mechanisms to gait variability in PD, reflecting the multifactorial nature of gait. Stride time variability and DLS time variability reflect different aspects of motor control, and appear to respond selectively to dopamine. Predictors of gait variability are selective to type of variability characteristic and medication status, which argues for closer examination of these outcomes.