Introduction

Walking in our natural environment can be considered a dual-task (DT) scenario that requires increasing cognitive resources with advancing age. Age-related decline of performance whilst walking in DT situations has been extensively investigated [1,2,3,4,5]. For instance, an age-related decline in gait performance has been observed when conducting arithmetic, memory or visual tasks concurrently with walking [5, 6]. Walking is not an automated task and requires structural and functional connectivity of neural brain networks. Changes in brain structure are common with ageing and require re-allocation of cognitive resources for fast and efficient operation of neural brain networks [7, 8] during complex activities. Higher age is further associated with reduced cognitive processing efficiency (e.g., decrease in nerve conduction speed and increased lateralization) [9], which is in turn associated with a decrease in cognitive performance such as diminished response time, working memory and processing of multiple tasks. These age-related cognitive changes affect daily-life task performance [10]. The level to which walking performance is affected by cognitive-motor interference is typically expressed as the dual-task cost (DTC). This is calculated as the percentage of decrements in performance in a dual- or multi-task relative to single task performance. It is proposed that, with advancing age, sensory and motor aspects of walking performance increasingly require cognitive control and attention. Several studies report a correlation between age-related declines in the sensory and motor system on the one hand and age-related declines in cognitive functioning on the other hand [11]. There is some evidence that decrements of gait performance in older people with a reduced postural reserve (motor abilities to maintain balance) can be independent of the cognitive performance [12]. Other studies showed that impaired executive function and attention affect walking performance of older fallers independent of physical ability [13, 14].

DT paradigms have become prominent to understand cognitive-motor interference (CMI) while walking in old age. These dual-task experiments have demonstrated that the extent to which the cognitive demand affects walking performance is exacerbated in old age [15], people with high risk for falls [16] and people with concerns about falling [17]. People’s tendencies to change their gait patterns during complex activities might result in an increased risk of falling [10]. Many studies reported more pronounced impairments of spatiotemporal gait parameters under dual-task conditions (including gait speed, step length, step width and double support time) in fallers compared to non-fallers [18,19,20]. Cognitive-motor interference in combinations with poorer physical abilities may increase a person’s risk of falling even further, especially in situations that require the adoption of a faster gait speed [21]. This is further impacted by poorer judgement of physical abilities, which has been linked to more collisions with oncoming cars in virtual reality experiments [22, 23]. The understanding of cognitive-motor interference in people with high fall risk or concerns about falling during walking under different cognitive dual-task conditions is still quite limited. Moreover, there is little information about which motor and cognitive task combinations require the highest attentional demands in older people and which mechanisms lead to insufficient resource allocation.

Theoretical models to explain cognitive-motor interference

Several theoretical models have been proposed to explain reduced walking performance in dual-task situations. The central bottleneck theory states that due to an information processing bottleneck only one task can be processed at a time; processing of a second task cannot commence until the first is complete. This bottleneck usually results in a longer response time for one of the two tasks [34,35,36]. The 4-dimensional multiple resource model [37] proposes that there will be greater interference between two tasks that utilise similar resources. Finally, the attentional resource theory suggests that declines in performance under DT conditions result from interference caused by competing demands for attentional resources, resulting in less attention available to each task [38, 39].

The attentional resource theory might especially apply to people with CoF. CoF is very common in older people and can lead to self-induced restriction of physical and social activities. In its most severe form, it can result in a persistent and dysfunctional disruption of attention. People with higher levels of CoF have difficulties to inhibit or ignore irrelevant information of the environment in the process of balance control. Therefore, CoF may compete for the limited resources of attentional focus to maintain balance control during complex activities [40] resulting in instability and increased fall risk. A meta-analysis by Ayoubi et al. [41] revealed that CoF is associated with increased gait variability during normal walking. This effect is amplified under DT conditions, due to reduced gait speed and step length (often referred to as cautious gait), especially in older people who also reduce their daily physical activity due to their CoF [42].

Performance is expected to deteriorate in complex situations if there are fewer resources available for performance than are required. Navon [43] defined resources as any internal input that is essential for processing and is available in limited quantities at any point in time. Walking requires coordination of peripheral sensory and neuromuscular systems, with higher-level cognitive processing, which gradually decline with age. It is therefore not surprising that with advancing age, cognitive-motor interference becomes more pronounced when performing complex daily activities [10, 36, 44]. Each task requires a reweighting of sensorimotor information depending on the requirements of the additional task [45]. When the sensory system delivers conflicting information, vision will dominate spatial processing, which impacts a person’s ability to coordinate sensory and cognitive processing to main upright [45]. In addition, studies indicate that increasing difficulty levels (from DT to multitask-performance or with different task complexities e.g. from processing speed to decision-making tasks; see Table 1) further amplify the effects of cognitive-motor interference on walking performance [46,47,48,49,50,51]. Systematic reviews have further highlighted that cognitive-motor interference rises based on the task domain and the individual’s abilities and resources [52, 53]. More specifically, tasks including controlled processes or motor components showed more decrements in DT performance of older people.

Table 1 Proposed taxonomy for cognitive dual tasks
Full size table

However, activities that heavily rely on postural control occasionally lead to an improved motor performance when combined with a secondary task [54]. The U-Shaped Non-Linear Interaction Model postulates that, depending on the complexity of the secondary task, motor and balance performance can increase or decrease [55]. For example, there might be a reduction of postural sway as a result of muscle co-contraction while concentrating on the cognitive task [56, 57], whereas postural sway may increase without additional cognitive performance with a secondary task [58]. The Supra-Postural Task Model [59, 60] provides additional details to explain the U-shape relationship between postural control and balance. The theory suggests that in specific situations the motor performance is necessary to reach the goal of the cognitive task (e.g. standing still to read a sign). In contrast to the U-shaped model, in the Supra-Postural Task Model effects are explained by situation awareness and not by task complexity [61].

Finally, the Task Prioritization Model [62] accounts for the strategies that an individual might use during complex activities. It postulates that older people are more likely to prioritize motor performance under threat of a loss of balance [63, 64]. This prioritization reduces the cognitive-motor interference and allows for reorganization of the cognitive-motor resources [65] to reduce the risk of falling. However, if the environment poses too many challenges (e.g. elevated surface), task prioritization is not always effective. Yogev-Seligmann and colleagues [66] found that older people with adequate balance abilities and capacity to identify hazards are able to focus on cognitive performance as long as balance is maintained. On the other hand, fallers are not able to shift attention in these situations [67], which could be explained by the impact of poor executive function and attention on walking performance of older fallers [13, 14].

Objectives

The primary objective of this review was to use a taxonomy for classifying cognitive tasks to gain insight in cognitive-motor interference within the study of falls in older people. Previous reviews concluded that gait speed under DT conditions is equivalent to gait speed as a single task in the prediction of future falls in older people [50, 68]. However, without a clear taxonomy of cognitive dual tasks, these conclusions might be premature. In addition, little is known about the effects of dual-task settings on older adults with CoF. A clear taxonomy will allow a better understanding of how cognitive-motor interference during complex activities is related to fall risk and concern about falls.

Methods

Search strategy

Databases were systematically searched by using OvidSp to search in Medline (1946 to 2019, Week 20), Embase (1974 to 2019, Week 20) and PsycINFO (1806 to 2019, Week 20). The search within the databases was limited to the English and German language. In addition, the reference lists of included articles were searched manually. Two reviewers (BW, MW) independently searched within titles and abstracts to identify all potentially eligible studies. Afterwards, these two reviewers independently assessed full paper copies of the identified potentially eligible studies to determine the studies to be included. Any disagreement on inclusion was resolved by discussion and through arbitration by a third reviewer (KvS, KD).

Inclusion and exclusion criteria

The inclusion criteria were: (i) older adults ≥ mean age of the sample was 60 years with a previous fall or CoF, (ii) the dual-task paradigm was used to discriminate fallers from non-fallers or people with high concerns about falling from people with low concerns about falling, (iii) utilized straight over ground walking at self-selected speed as the primary motor task, (iv) reported gait measurements during both single and dual-task performance, or the effect of dual-tasking on gait performance (more than one gait cycle), (v) clear description of the dual-task situation, (vi) reported adequate data to calculate effect sizes either from descriptive or inferential statistics, (vii) interventional studies were included if the effect of dual-tasking on gait at baseline was reported. The exclusion criteria included: (i) population with brain injuries or diagnosed cognitive decline, (ii) physical impairments (e.g. using a cane or walker) and (iii) chronic diseases (e.g., multiple sclerosis or Parkinson’s disease). Moreover, studies with a secondary analysis of previous reported results were also excluded.

Selection criteria

Studies comparing fallers and non-fallers were included if the method section reported of the number of falls. Prospective studies were considered if they compared fallers and non-fallers at baseline (retrospective) or at the follow-up measurement und ST and DT conditions.

Studies addressing CoF were included if they classified the participants according to the “falls efficacy scale international (FES-I)” [69] score, the activities-specific balance confidence (ABC) scale [70] or if they asked the participants using a single item question if they were afraid of falling during activities of daily life.

Studies that included walking under DT conditions were included. This includes studies that investigated at least one walking task (in a DT setting; according to the definitions of spatiotemporal gait parameters addressed in Table 2), studies that compare ST and DT performance, and studies that investigated DT performance in healthy or balance-impaired (fallers) older adults in either a randomized control trail (RCT), an experimental-control group design or an old-young comparison. Moreover, studies with a secondary motor task were also included. Additionally, every concurrent task was assigned to a “stimulus-response-condition” (visual-verbal, visual-manual, auditory-verbal, auditory-manual) and classified according to our taxonomy of cognitive tasks (see Table 1).

Table 2 Spatiotemporal gait parameters
Full size table

Quality assessment

Quality assessment of the included articles was based on the Standard Quality Assessment Criteria (SQAC) for evaluating primary research papers proposed by the Alberta Heritage Foundation for Medical Research [71]. As the review did not focus on RCTs, the quality criteria for RCTs were not assessed. The quality criteria, as described in SQAC, were: (1) sufficient description of the question/objective; (2) appropriate study design; (3) appropriate method of participant selection or source of information/ input variables; (4) sufficient description of participant characteristics; (5) report of means of assessment with outcome measures well defined and robust" to measurement or misclassification bias (6) appropriate sample size; (7) appropriate analytic methods and method description; (8) report of estimate of variance in main results; (9) control for confounding; (10) sufficiently detailed report of results; and (11) conclusions supported by the results.

Participant selection was verified by comparing the sample with the conclusions drawn from the experimental results. A full point for appropriate sample size was given when either an a priori calculation of sample size had been described or the sample size was a full cohort. Based on the analytic methods employed (8), important statistical values (according to the APA-Manual [72]) had to be included to obtain a full quality score. BW and MW or KvS performed the assessment independently and the results presented in Table 3 were concurred on. Each criterion scored one point if partly fulfilled and two points if completely fulfilled. Points were added up and resulted in the quality score. The necessary score for a study of high quality was defined to be 17 out of 22 (75%) and 10–16 points for standard quality according to the SQAC. No point was given if general remarks had to be made (indicated by brackets; Table 3). Moreover, we reported some general methodological issues (cf. column general marks). Studies were included in the meta-analysis if they had quality score of 7 or more.

Table 3 Quality score
Full size table

Data extraction

Table 4 provides an overview of all included studies including the authors, year of publication, study design and aims, population with discrimination to fallers/non-fallers or participants with concerns or no CoF, observed walking parameters and description of the DT setting. The main results of the studies were extracted to Table 5. This includes task order, outcome measures used to assess and report the concurrent tasks performance and instructions given to participants, and study results. Data were recorded as a mean and standard deviation (SD) if reported, with sample size and number analyses in each group (fallers vs. non-fallers or participants with concerns or no CoF).

Table 4 Included studies with fallers
Full size table
Table 5 Data extraction fallers/ non-fallers
Full size table

Statistical analysis of the meta-analysis

For each of the outcome variables of interest (gait speed, cadence, stride length, step length; see Table 2) we collected the gait data for single and dual-task performance. The gait data was presented as differences in means (MD), since the outcome measurements were made or could be converted on the same scale (e.g., meters per seconds). Most of the studies reported means and SDs permitting effect size estimation, otherwise, they were derived from other summary statistics reported in the articles, such as t-values or p-values. The gait data from individual studies were then pooled in meta-analyses to estimate the overall effect of cognitive-motor interference of gait. Studies were grouped by cognitive task domain and individual meta-analyses were conducted for each outcome: gait speed, cadence, stride length and step length.

In order to determine whether studies shared the same overall effect size or whether the overall effect for a given outcome was modified by certain factors, we conducted a subgroup analyses on studies that directly compared two factors of interest (e.g., arithmetic task vs. verbal fluency tasks) or two groups of participants (e.g., fallers vs. non-fallers) within the same study. Subgroup analyses were conducted using a mixed-effects model and the summary effects within subgroups were computed using a random-effects model. Moreover, to further analyse the differences between fallers and non-fallers as well as participants with and without CoF, DTC were calculated by subtracting the DT values from the ST values. A random-effects model with a generic inverse variance method was used in the pooled analyses, which gives more weight to studies with less variance. Results are presented as effect size with 95% confidence interval (CI) and respective values for null hypothesis tests (e.g., cognitive-motor interference has no effect on gait). Heterogeneity between studies was investigated by calculating the Q-value and I2 statistic which quantified the proportion variation that is due to heterogeneity rather than chance. Quantitative syntheses and meta-analyses were produced using Review Manager 5 Software (RevMan 5).

Results

Databases and references identified 2,670 unique articles for consideration. After abstract consideration and title screening, a total of 71 studies were included for further consideration. Reasons for exclusion were studies using participants with neurological disease (e.g., Multiple Sclerosis, Stroke), studies using obstacle negotiation or Reviews. After applying the inclusion criteria, 19 studies were assessed for quality and 16 papers were included in the meta-analysis (cf. Fig. 1; for excluded studies cf. Table 6 and Table 7).

Fig. 1
figure 1

Flow chart of the systematic review procedure

Full size image
Table 6 Excluded paper
Full size table
Table 7 Excluded paper meta-analysis
Full size table

Thirteen studies showed high quality scores (> 16) and seven studies were of good quality (according to [71]). The study by Yamada et al. [86] was excluded due to a quality score < 10. Table 4 gives an overview of all included studies addressing the comparison of fallers vs. non-fallers and participants with and without concerns about falling. The study by Wollesen et al. [90] could not be integrated into the meta-analysis because they used a fixed gait speed in their measurement design.

Fallers vs. non-fallers

Description of the included studies comparing fallers and non- fallers (N = 15)

The mean age of the study population was between 67 years [21, 84, 85] and 87 years [19]. The sample sizes of the studies varied between N = 16 [84, 85] and N = 1350 [78].

Five studies included a prospective design [19, 74, 76, 77, 85].

The included studies used the following dual-task settings:

  • Arithmetic tasks: n = 7 studies used counting backward tasks [19, 20, 74, 75, 80,81,82], conducted as counting in steps of one (n = 3), three (n = 3) or seven (n = 3) (cf. Table 3).

  • Verbal fluency tasks: n = 7 studies used verbal fluency tasks [20, 21, 75,76,77, 80, 81]

  • Motor tasks: n = 5 studies used a motor task [20, 21, 80, 83, 85]

  • Other tasks: visuo-spatial task [20], Stroop task [20], listening and memory task [82] and reciting of letters of the alphabet [85].

  • A total number of six studies analysed more than one task [20, 21, 75, 80,81,82].

Overall, the studies comparing fallers and non-fallers examined 32 different gait quality variables. Gait speed or velocity was assessed by n = 14 studies [19,20,21, 74, 75, 77,78,79,80,81,82,83,84,85]. Other gait measures included duration to walk a defined distance (n = 2) [19, 77], step length (n = 3) [21, 80, 85], stride length (n = 4) [14, 83,84,85], cadence (n = 6) [19, 21, 77, 83,84,85], step time (n = 3) [80, 83, 85], stride time (n = 5) [21, 77, 81, 83, 85] and double support time (n = 3) [77, 80, 85]. Several studies used gait parameters of variability (n = 14; eg.: stride time variability (n = 3), gait speed variability (n = 2) and swing time variability (n = 2)). In addition, some studies focused on Center of pressure (CoP) or Center of mass (CoM) displacements, or mechanical power in anterior (AP) and medio-lateral (ML) direction during gait cycles. These outcomes were not included into the meta-analysis because of lack of consistency in calculation methods among studies or infrequent use. To measure gait characteristics, a stopwatch (n = 6; from 10 m up to 30 m distance), the GAITrite rite system or another electronic walkway (n = 8; from 8 m up to 12 m), camera systems (e.g., Vicon n = 3) or insoles (e.g., F-Scan n = 3) were used.

Fig. 2
figure 2

Forest plot meta-analysis of ST performance between non-fallers and fallers

Full size image

Differences on cognitive-motor dual task performance between non-fallers and fallers

Four studies could not be integrated into the meta-analysis because the mean values and SD for the analysed gait data were not reported in comparison of non-fallers and fallers and unavailable after attempting to contact the authors [76, 78, 81]. Independent of the task settings, there were no differences of the gait decrements under DT conditions between fallers and non-fallers (cf. Table 5). Mostly, fallers showed reduced performance of the spatiotemporal gait parameters in comparison to non-fallers. Only two studies used a coefficient of variation [81, 82] and revealed significant differences between fallers and non-fallers with increased variation in fallers. Reelick [81] found a significantly reduced walking performance for the verbal fluency task in comparison to the arithmetic task. Nordin et al. [80] also revealed differences for their task conditions; gait speed increased for the motor-tasks (carrying a cup or a tray) and gait speed decreased for the cognitive conditions (verbal fluency and counting backwards) fallers compared to non-fallers.

Results of the meta-analysis fallers vs. non-fallers

The forest plot of Fig. 2 shows significant mean difference of 3.32 [95% confidence interval 0.66–5.99] between non-fallers and fallers for ST gait speed with reduced performance for fallers. However, these results were heterogeneous (I2 = 39%; cf. Fig. 2). There were no effects for step length or stride length. Under DT conditions, fallers had a reduced gait speed in comparison to non-fallers with a mean difference of 6.10 [2.23–9.98] (I2 = 44%; cf. Fig. 3).

Fig. 3
figure 3

Forest plot meta-analysis of dual-task effect on gait different gait measurement between non-fallers and fallers

Full size image

Figure 4 repeats the findings for gait speed under ST and DT conditions and shows the mean difference in DTC (defined as DT minus ST). The meta-analysis showed that there were higher decrements in gait speed for fallers in comparison to non-fallers under DT conditions. However, if the DTC were calculated (Fig. 4), there were no reduced DTC observed for non-fallers.

Fig. 4
figure 4

Comparisons of ST and DT gait speed and resulting dual task costs (DTC)

Full size image

Figure 5 visualizes the DTC for the different cognitive task domains. Increased DTC for fallers compared to non-fallers could only be observed for verbal fluency and motor dual-tasks but failed to be significant. The overall effect of the different task conditions was also not significant.

Fig. 5
figure 5

Comparisons of ST and DT and resulting DTC for the different task conditions

Full size image

Participants with concerns about falling vs no concerns about falling

Description of the includes studies (N = 4) comparing participants with CoF

The mean age of the study population was 69.8 years [90] up to 80.6 years [89]. Sample sizes varied between N = 85 [90] and N = 1307 [88]. The included studies used different dual-task settings:

  • Arithmetic tasks: The study by Reelick [100] used a counting backward tasks (subtracting 7 s) and the study by Asai [87] used a counting backward task (subtracting 1 s) (cf. Table 4).

  • Verbal fluency tasks: Donoghue et al. [88] (recite alternative letters of the alphabet) and Reelick et al. [89] (naming animal species as much as possible) used a verbal fluency task.

  • Other tasks: The RCT by Wollesen et al. [90] was conducted with a visual-verbal Stroop task.

Studies comparing participants with and without CoF examined 16 different gait variables (cf. Table 5); i.e. gait speed (n = 3), stride time variability (n = 1), step width (n = 2), step length (n = 1), stride length (n = 2). Two studies used different variability calculations (n = 2). Moreover, two studies [87, 89] focused on CoP or CoM displacements in AP and ML direction during gait cycles. To measure gait performance, the GAITrite system or another electronic walkway (n = 2; from 5 m up to 10 m), a triaxial accelerometer (n = 1) or a treadmill (n = 1) were used (cf. Table 5).

Differences on cognitive-motor-performance between participants with and without concerns about falling

As reported in Table 5 participants with and without CoF showed comparable DTC. Moreover, all studies showed that participants with CoF had a poorer walking quality (e.g., reduced walking speed with accompanying step length or increased variability) in the ST condition compared to people without CoF. With regard to the different task settings, the two studies that examined two different cognitive dual-tasks found different reactions in all participants according to the task. The study of Asai et al. [87] analysed an arithmetic DT situation and a motor-motor DT situation; and found that both tasks resulted in reduced walking speed. The motor-motor DT resulted in reduced (and therefore improved) body sway in ML and AP direction in comparison to the arithmetic DT situation. Reelick et al. [90] investigated an arithmetic DT situation and a verbal fluency task, and found no task differences. The meta-analysis revealed a significant difference of gait speed between participants with and without CoF under ST (mean difference: 12.41 [9.97–14.84]) and DT (mean difference: 10.61 [7.58–13.40]) conditions. The differences for the DTC failed to show significance (mean difference: 1.63 [− 1.01–4.27]; cf. Fig. 6).

Fig. 6
figure 6

Comparisons of ST and DT and resulting DTC for participants with and without concerns about falling

Full size image

Discussion

The aim of this systematic review and meta-analysis was to provide a taxonomy of different dual-task settings and test their relations to cognitive-motor decrements with fall risk and CoF. Additionally, the cognitive tasks were regarded separately with the purpose to find a dual-task taxonomy or classification of the DT settings that are most beneficial to identify cognitive-motor interference in older fallers and older people with CoF.

Differences of DT performance on spatiotemporal gait parameters between non-fallers and fallers

The results of the meta-analysis suggested that gait speed and cadence in ST and DT conditions can discriminate between fallers and non-fallers. Studies classifying people as fallers and non-fallers were based primarily on retrospective falls, with only two studies being prospective [19, 96]. These results confirm previous systematic review evidence which showed differences in gait speed between fallers and non-fallers [50, 68]. With regard to the associated DTC, only five of eleven studies found higher decrements in gait speed from ST to DT for fallers in comparison to non-fallers (Fig. 4). The overall DTC failed to be significant between these two groups in our meta-analysis. There were only small amounts of DTC for both groups and the standard deviations were large. In line with the results of other studies that could not be included in the meta-analysis, fallers and non-fallers both show decrements in gait speed in ST and DT conditions (cf. Table 5 and Fig. 4). These decrements are not significantly different between groups which is inconsistent with the hypothesis that non-fallers and fallers differ in their ability of task prioritization [16, 67]. Fallers walk significantly slower than non-fallers in ST conditions; however, step length and stride length, which are known to be highly correlated with gait speed [91], did not differ significantly between groups. Specific recommendations on whether or not cognitive-motor-interference increases fall risk cannot be provided. These results confirm the findings by Zijlstra et al. [68] and Menant et al. [50] who also reported no additional benefit of DT walking as a measurement to discriminate fallers from non-fallers. Nevertheless, it is important to note that gait performance includes different components of functional performance such as maximal walking velocity, gait economy, walking effectiveness, efficiency and safety. These aspects might be more relevant to estimate fall risk. Therefore, future studies should address these components of gait performance in tailored DT settings.

Differences of DT performance between participants with and without CoF

People with CoF showed greater gait decrements under ST and DT conditions compared to people without CoF. The overall effects from the meta-analysis suggested that the effects of CoF were larger (11.61; CI: 9.75–13.48) in fallers compared to non-fallers (4.12; CI: 2.20–6.03). CoF is common in people with and without a previous fall history and the prevalence rates are higher than falls themselves [93]. It has been suggested that people with CoF have difficulty inhibiting or ignoring irrelevant information of the environment when controlling their balance in complex and DT situations [40]. Many daily life activities include some level of dual-tasking in which executive functioning or performance (i.e. inhibition) are required. CoF might compete for these limited resources of attentional focus to maintain their balance [52], which would result in a more pronounced slowing of their walking speed under DT conditions (cf. Fig. 6) in people with CoF irrespective of their fall history or fall risk. However, our analyses were not able to confirm this hypothesis as DTC was not significantly different between people with and without CoF.

Influence of the task condition

A large variety of cognitive tasks have been used to assess cognitive-motor interference in the literature. As part of this review, a total of 11 different DT-conditions were used to compare non-fallers and fallers on DT walking performance (Fig. 5). According to the proposed taxonomy (Table 1) mental tracking tasks, especially counting backward tasks by numbers in 1 s, 3 s or 7 s are the most commonly used task sets. Overall, we were able to compare three types of cognitive dual-tasks (i.e. arithmetic, verbal fluency and motor tasks) within the meta-analysis of this review. Two of them belong to the same category of our taxonomy (mental tracking, cf. Table 1). The third one included an additional motor task. However, all task settings affected DTC similarly, and the pooled effect (mean difference: − 1.00 [− 3.72–1.73]) had low heterogeneity (I2 = 0%).

Other cognitive tasks such as reaction time and decision making tasks for processing speed and controlled processing tasks, [92] were not integrated in the task setting of the included studies but could be relevant for navigating in daily traffic situations. In addition, previous studies have suggested that more complex tasks such as working memory tasks, discrimination tasks or visuospatial tasks would have a greater impact on the DTC (for an overview see Lacour et al. [52]) but this could not be confirmed by this review due to the limited studies using these tasks. Furthermore, within the available data there were also no marked differences between the different types of cognitive tasks. On the other hand, there is evidence that mental tracking tasks like verbal fluency tasks increase the DTC more significantly for fallers compared to non-fallers [81], due to the additional load on the working memory for these tasks. However, this review was not able to confirm this hypothesis. Finally, motor-motor DT condition also did not show significant differences in DTC between non-fallers and fallers. Both studies by Toulotte et al. [83, 84] suggested a more pronounced DTC when carrying in glass of water, suggesting this would slow participants down as they need to observe the glass of water in their hand. However, other studies have suggested the opposite [80], as a result of a forward flexion of the trunk when carrying a tray with a glass of water in front of the body.

Implications of the results

Similar to previous reviews, we were not able to confirm differences between fallers and non-fallers in DTC. One reason for this result might be that we were only able to compare three types of dual-task settings (i.e. arithmetic, verbal fluency and motor tasks) within the meta-analysis. Therefore, additional studies are required to examine the discriminatory ability of walking performance with and without concurrent reaction time, controlled processing, visuospatial, working memory and discrimination tasks. Study designs comparing different DT-settings in smaller samples [20] or randomised trials with a representative larger sample size could be used to systematically address different cognitive processes and their complexities. In addition, it might be important to consider an individual’s biography before deciding on a DT. One might argue, that a maths teacher might find a counting backwards task more intuitive, while a librarian might be more comfortable with verbal fluency tasks. More work is required to test this hypothesis. Tasks that include visuo-spatial information processing or higher executive functions (e.g., inhibition within a Stroop-task) [2] might have greater potential in discriminating between fallers and non-fallers. These tasks may be less dependent on people’s biography. However, these task-settings might be difficult to use in clinical settings and with short walking distances. In addition to the cognitive dimensions of the task settings, the walking conditions and parcourse need to be reflected, as a straight walking course does not sufficiently address real-life gait. The ongoing development of wearable technology might be one solution to overcome measurement set up problems.

Limitations

Overall, the quality of the included studies was good. Nevertheless, there are some issues that need to be discussed. First, spatiotemporal gait parameters were assessed using diverse measurement methods, varying between the crude use of a stopwatch to accelerometers and electronic walkways [94]. Second, there is not a common length of the walking tracks with many studies using distances that are too short to see a DT effect. According to Lindemann et al. [95], the distance to achieve a steady walking state increases with higher gait speed. Third, studies report different spatiotemporal gait parameters. Especially, spatiotemporal gait parameters related to balance, such as step width, double support time, gait stability and variability, were not reported frequently enough to be included in the meta-analysis. It is possible that the effect of DTC would be visible on such measures before it affects gait speed especially over short distances. Fourth, the short distances might influence prioritisation of the motor and cognitive tasks. The short distances also limit the time available for the cognitive dual-task, which might explain why the meta-analysis could not show a different cognitive-motor interference on gait between fallers and non-fallers. Finally, most of the studies did not report the motor and the cognitive DTC. This means that there is no control for the attentional focus of the participants, rendering it unclear if the performance decrements result from the attentional focus or from cognitive-motor interferences. Finally, to gain information about the influence of the DT taxonomy on DTC, this review integrated only studies with straight walking. This was necessary to overcome the problem that gait execution while changing directions, walking in curves or reacting to external perturbation, has a different impact on spatiotemporal gait parameters as well on the cognitive performance.

Conclusions

Overall, the large diversity of studies and types of cognitive dual-tasks do not allow us to provide conclusive recommendations for clinical testing of cognitive-motor interference while walking. In agreement with previous studies [50, 78], we found no additional benefit of DT gait analysis to differentiate between fallers and non-fallers. Similar results were found when comparing people with and without CoF. However, our analyses also reveal that several domains of cognitive dual-tasks have not yet been investigated. The proposed cognitive task taxonomy will assist in systematic assessment of these tasks and their effect on gait.