How Many Trials Are Needed in Kinematic Analysis of a Reach-to-Grasp Task?-a Study in Persons With Stroke and Non-Disabled Controls


 Background Kinematic analysis of the 3D reach-to-grasp drinking task is recommended in stroke rehabilitation research. The number of trials required to reach performance stability, as an important aspect of reliability, has not been investigated. Thus, the aims of this study were to determine the number of trials needed to reach within-session performance stability and to investigate trends in performance over a set of trials in non-disabled people and in a sample of individuals with chronic stroke. In addition, the between-sessions test-retest reliability in persons with stroke was established. MethodsThe drinking task was performed at least 10 times, following a standardized protocol, in 44 non-disabled and 8 post-stroke individuals. A marker-based motion capture system registered arm and trunk movements during 5 pre-defined phases of the drinking task. Intra class correlation statistics were used to determine the number of trials needed to reach performance stability as well as to establish test-retest reliability. Systematic within-session trends over multiple trials were analyzed with a paired t-test. Results For most of the kinematic variables 2 to 3 trials were needed to reach good performance stability in both investigated groups. More trials were needed for movement times in reaching and returning phase, movement smoothness, time to peak velocity and inter-joint-coordination. A small but significant trend of improvement in movement time over multiple trials was demonstrated in the non-disabled group, but not in the stroke group. A mean of 3 trials was sufficient to reach good to excellent test-retest reliability for most of the kinematic variables in the stroke sample. Conclusions This is the first study that determines the number of trials needed for good performance stability (non-disabled and stroke) and test-retest reliability (stroke) for temporal, endpoint and angular metrics of the drinking task. For most kinematic variables, 3 trials are sufficient to reach good reliability. This knowledge can be used to guide future kinematic studies.


Background
Analysis of multi-joint 3D kinematics is needed to understand the underlying mechanisms of the altered movement strategies commonly seen post stroke (1). Unlike traditional clinical assessments, objective measures of movement quality allow differentiation between behavioral recovery and compensation in evaluation of treatment effects (2)(3)(4). Here, the kinematic analysis can provide detailed and objective information about movement performance and movement quality during everyday activities, such as reach-to-grasp (5,6).
Reach-to-grasp is frequently used in daily activities and its performance in non-disabled individuals is characterized by e cient spatiotemporal coordination of the arm and hand segments for transport and grasping (7). Regaining arm-and hand function post-stroke is one of the highest priority goals in rehabilitation, and still about 65% of the patients with hemiparesis have impaired ability to reach, grasp and handle objects at 6 months after stroke onset (8). Motor performance of reach-to-grasp tasks in the stroke population shows longer movement time, lower peak velocity, decreased elbow extension, greater arm abduction and trunk displacement, and decreased smoothness as compared to non-disabled controls (5,(9)(10)(11). Among the reach-to-grasp tasks, drinking from a glass has, due to its ecological validity and ease of standardization, been recommended as a functional task for quantifying quality of movement in stroke rehabilitation research (12).
Another aspect that needs to be considered in performance of daily purposeful tasks is variability of movements. Variability is inherent in human movement control, i.e. different neuromotor processes are available to produce automatic movement strategies needed for achieving goals in daily life (13). The concept of movement variability is de ned as typical variations in motor performance when a task is repeatedly being executed (14), which is something that needs to be taken into account when conducting clinical research studies. Optimal movement variability is crucial for healthy motor control (13,15). A high level of automaticity and relatively constant variability is, however, expected when a well-known activity is repetitively performed (16).
Requests for standardization of kinematic analysis of upper extremity movements have been highlighted (11) and for research purposes several efforts have been made to agree on which tasks to study and which systems and metrics to use (5,(9)(10)(11)(12). Clinimetric properties, including reliability, validity and responsiveness, have been reported for some kinematic metrics (9,11,17,18) although more studies are needed (19,20). One aspect of reliability that has been sparsely investigated is the performance stability of selected variables within a session of a series of trials. Most of the studies of reach-to-grasp tasks in stroke populations include 3-10 trials per task although in few studies up to 20 trials have been reported (5,11). A recent consensus on kinematic studies in stroke recommended at least 15 trials to be collected, both for 2D performance assays and 3D functional tasks (12).
Hence, the question of how many trials that are needed to reach performance stability of kinematic measures in goal-directed reach-to-grasp tasks remains. A study in non-disabled subjects, using kinematics from an optoelectronic system, de ned the minimum number of trials needed to reach su cient performance stability for trajectory analysis of discrete pointing movements to be e.g. 3 for movement time and peak velocity whereas constant error required 47 trials (21). A study in persons with subacute stroke, where also 3D motion capture was used, reported that 5 trials was su cient to get reliable results for reaching kinematics (22).
To our knowledge, no studies have de ned the number of trials needed to achieve performance stability, i.e. good reliability, in kinematic measures of goal-directed reach-to-grasp tasks, nor has this been investigated in people with disabilities. Thus, the primary aim of this study was to determine the number of trials needed to reach good performance stability of the kinematic variables during the drinking task in non-disabled people and in a sample of individuals with chronic stroke. Further, the performance stability over the set of multiple trials was investigated. In addition, the between-sessions test-retest reliability of selected kinematics in a sub-sample of individuals with stroke was established.

Participants
This study included 44 non-disabled participants who were recruited through personal contacts and general advertisements during 2016-2019 in the urban area of Gothenburg in Sweden.
The non-disabled participants were included when they were between 30 and 85 years, had not being diagnosed with any medical condition that would potentially in uence the movements of the upper extremity or upper body, and perceived themselves as healthy.
Potential participants were excluded, if they showed any observable neurological signs (e.g. tremor), di culties to follow simple instructions or had uncorrected visual acuity that in uenced the movement performance. The non-disabled participants performed the kinematic drinking task at one occasion.
In addition, eight participants with stroke, screened for separate single case design studies between 2018 and 2020 were included. Inclusion criteria were a diagnosis of stroke at least 6 months earlier, ability to adhere to the upper extremity virtual reality intervention study protocol requiring ability to hold an object like remote control with the more-affected hand, and able to attend the physical visits over 15 weeks' time at the research site (23). For the current analysis, only data from the stable phase (phase A) prior intervention was used.
Five participants with stroke had kinematic data available from four separate testing sessions (with one week apart), and three had data only from one screening session.
Background data on age, sex, hand dominance, body height and weight were registered for all participants. The type and side of stroke and time since onset were also recorded for participants with stroke. Upper extremity motor impairment in stroke was assessed with the Fugl-Meyer Assessment of Upper extremity (FMA-UE) (24,25) and the activity limitation with the Action Research Arm Test (ARAT) (26,27). In addition, the non-motor domains of the FMA-UE (sensation, range of motion and pain) and muscle tone (modi ed Ashworth Scale) (28) for elbow and wrist joint movements were assessed. The demographic and clinical characteristics of all participants are shown in Table 1. and oral and written informed consent was obtained from all participants.

Kinematic movement analysis
The standardized established kinematic analysis testing protocol for drinking task was used (10,12,17).
Kinematic data was acquired with a 5-camera high speed optoelectronic motion capture system (Prore ex MCU240 Hz, Qualisys AB, Gothenburg, Sweden). The cameras emit infra-red light that is re ected by the circular markers placed on anatomical landmarks on the body. The eight markers (12 mm) were placed on the tested hand (III metacarpophalangeal joint), wrist (styloid process of ulna), elbow (lateral epicondyle), on both shoulders (acromion), trunk (sternum), forehead and the drinking glass. Kinematic data was ltered with 6-Hz second-order Butterworth lter in forward and backward direction and analyzed off-line in the Matlab software (R2019B, The Mathworks Inc).
The drinking task was divided into 5 phases: (1) reaching to grasp the glass, (2) forward transport of the glass to the mouth, (3) drinking a sip of water, (4) transporting the glass back on the table, and (5) returning the hand back to the starting position.
For the standardization of the sitting position, the chair and table height were adjusted to attain 90° knee and hip exion, 90° elbow exion while the upper arm was in vertical and forearm in horizontal position (17). The wrist was aligned with the table edge with the palm resting on the table. A hard-plastic drinking glass containing 100 ml water was placed 30 cm from the table edge (approximately 75-80% of the arm's length) in the midline of the body.
The trunk was not restrained, although the participants were instructed to sit with their back against the back of the chair. After few familiarization trials, ensuring that the participants had understood the instructions correctly, the drinking task, including all 5 phases, was repeated in self-paced natural speed at least 10 times unimanually, starting with the dominant or less-affected arm. The rest between each trial was approximately 5 seconds.
A set of kinematic variables describing both temporal and spatial characteristics of the movement performance, including end-point, angular and displacement variables, were obtained for the analysis.
De nitions of the kinematic variables are provided in Table 2. Time is calculated for the entire drinking task and separately for each phase. The start and end of the movement was de ned as the point in time when the velocity exceeded or was below 2% of the maximum velocity in the reaching or returning phase, respectively. Detailed de nitions for each phase are available in a previous publication (Alt Murphy et al. 2018 Movement units were computed from the tangential velocity pro le separately for rst two movement phases (reaching and forward transport), last two phases (back transport and returning) and as a summed total of these four phases (NMU total). One movement unit was de ned as a difference between a local minimum and next maximum that exceeded the amplitude limit of 20 mm/s, minimum time between two subsequent peaks was set to 150ms. NMU indicates movement smoothness. NMU phase 1&2

NMU phase 4&5
Peak velocity (mm/s) Peak tangential velocity of the hand marker in the reaching phase. Kinematic data from 10 trials was available for 68% and 78% of the non-disabled and stroke participants, respectively. All remaining sessions had 9 successful trials. Hence, in the analysis of performance stability, systematic trends and test-retest reliability 9 trials were used. Three trials from two non-disabled participants showed distinctively lower values of the inter-joint coordination. A visual analysis con rmed that these deviating values were caused by a backward movement of the hand prior forward reaching and these trials were therefore excluded from analysis.
The performance stability was veri ed through analysis of reliability, i.e. the repeatability of the selected kinematic variables, based on the intraclass correlation coe cient, ICC. The ICC that represented the absolute agreement for measurements that are averages of k trials on randomly selected individuals (29) was used to determine the number of trials needed to reach performance stability for each variable.
The ICC values were calculated separately for the non-disabled participants and participants with stroke, but also for data from the two groups together. For the latter combined ICC scores, the non-dominant arms of the non-disabled participants and moreaffected arms of the individuals with stroke were used.
Thresholds for the ICC were set according to recommendations by Koo and Li 2016 (30), which are based on the 95% con dent interval of the ICC estimate. Values of ICC were interpreted as poor (less than 0.50), moderate (0.50-<0.75), good (0.75-0.90), and excellent (greater than 0.90).
In order to determine the number of trials needed to reach good reliability, a series of ICC was calculated for each variable, where each ICC in the series represents the ICC value based on n consecutive trials (n = 1,…, 9). The ICC that reached ≥ 0.75 gave the recommended number of trials for each variable.
The systematic within-session trend was investigated by comparing the average of trial 1-3 with the average of trial 7-9 from the same occasion. A paired t-test was used, and the signi cance level p ≤ 0.05 was used to reject the null hypothesis that no trend existed. To control for multiple comparisons, p values were adjusted with Holm's correction (31).
The test-retest reliability of kinematic variables was analyzed in a subset of ve persons with chronic stroke who had repeated the drinking task at four occasions with one week between each occasion. The measurements in persons with stroke were obtained during an assessment phase prior an intervention and were considered as stable. The test-retest reliability was analyzed by computing an individual average for each person, variable and occasion based on n trials (n = 1,…,9).
The ICC that represented the absolute agreement for single measurements was used (since the average computed for each occasion was de ned as a single measure) to determine the number of trials needed to reach good test-retest reliability for each variable in this subgroup. The same threshold levels were used as when analyzing performance stability, i.e. ICC ≥ 0.75 represented good test-retest reliability.

Results
Background characteristics of the participants are shown in Table 1.
There were no statistically signi cant differences between the non-disabled participants and individuals with stroke in terms of age, sex, body height and weight.
All participants were right hand dominant.

Performance stability
The values for all kinematic variables for dominant and non-dominant arms in non-disabled and for the more affected arm in persons with stroke are reported in Table 3. ICC values as a function of number of included trials needed to reach good performance stability of kinematic measures are shown in Fig. 1. Number of trials needed to reach good performance stability are summarized in Table 4. The combined ICCs (non-dominant arms of the non-disabled participants and more-affected arms of the individuals with stroke) revealed that 18 of 21 variables reached good to excellent reliability for averages based only on 2 to 3 trials. More trials were needed for Movement time (MT) reaching (4 trials), MT returning (8 trials) and Time to peak velocity (6 trials). In the analyses of the non-disabled group alone the results were similar except for Number of Movement Units (NMU, 3 to > 9 trials) and Inter-joint coordination (4 trials). Even when only 3 trials were needed for NMU total of the dominant arm to reach good reliability, 9 or more trials were required for NMU of the non-dominant arm. The between-individual variations for these variables were low in the non-disabled group compared to the participants with stroke (see standard deviations reported in Table 3). In the separate analysis with participants with stroke alone, more than 3 trials were needed for MT reaching (5 trials), MT returning (8 trials) and Time to peak velocity (> 9 trials).  The systematic within-session trends between the rst 3 trials (trial 1-3) and the last 3 trials (trial [7][8][9] are shown in Fig. 2. Small but signi cant trends (p < 0.001) were observed in movement time variables in the non-disabled group, while no trends were found in the stroke group.

Test-retest reliability in a subgroup of individuals with stroke
In the subset of ve participants with hemiparesis after stroke, 17 out of 21 variables showed good or excellent test-retest reliability if the average value from each occasion were computed from 2 to 3 trials ( Fig.   3 and Table 4). For MT returning > 9 trials were needed. For the Wrist angle variable, the ICC was close to 0.70 after 2 trials, but reached over the level of ≥ 0.75 after 6 trials. The reliability remained moderate for Time to peak velocity over the 9 trials and for Peak velocity the reliability remained poor (Fig. 3).

Discussion
This study determined the minimum number of trials needed to reach good performance stability of kinematic variables obtained during the drinking task both in non-disabled persons and in a sample of individuals with chronic stroke. The results revealed that for most kinematic variables only 2 to 3 trials were required to reach su cient performance stability. Small but signi cant trends were noted for shorter movement times in the non-disabled group for the last 3 trials compared to the rst 3 trials. In the stroke sample, a good to excellent test-retest reliability was reached for many variables when less than 3 trials from each occasion were used in the analysis. However, more trials were needed for movement time in reaching and returning as well as for wrist angle. Only moderate reliability was reached for the time to peak velocity and poor reliability was observed for the variable peak velocity in the stroke group.

Number of trials needed to reach good performance stability
The current study is the rst to demonstrate that only 2 to 3 trials are required to reach good performance stability for most kinematic variables of the drinking task. This nding was valid both for non-disabled and for stroke participants and is in line with two previous studies analyzing reaching kinematics using optoelectronic systems (21,22). Blinch et al. reported that not more than 3 trials were required to achieve good within trial reliability of movement time and peak velocity during fast visually guided pointing tasks in non-disabled participants (21). Likewise, Hansen et al. demonstrated that 5 trials were estimated to be the minimum number required to get reliable ICC estimates for most of the kinematics when reaching for low and high targets in persons with subacute stroke (22).
Similar results have also been shown with other measurement systems in non-disabled individuals. A study using a virtual reality gaming Kinect system showed that 2 to 5 trials during reaching were needed to achieve performance stability in movement time and elbow and shoulder range of motion (32). Additionally, when using an inertial sensor system, comparable results of 3 trials was considered enough to reach acceptable levels of reliability for movement time and shoulder and elbow range of motion during a drinking task in non-disabled participants (33). These results con rm that for most of the kinematic variables a set of 3 trials would be su cient. However, more trials in a range of 4-6 and ≥ 8 trials would probably be needed for certain variables and study groups (e.g.

non-disabled participants).
Even though the total movement time for the drinking task only required 2 trials to reach good performance stability, up to 5 trials were needed for movement time in reaching (stroke) and up to 8 trials for movement time during returning (stroke and non-disabled). Post-stroke, abnormal muscle activation synergies and inadequate inter-joint coordination have been suggested to be the prime contributing causes to reaching dysfunction (10,34,35). In addition, abnormal inter-segmental dynamics, particularly regarding suppressed interaction torque and de cient feedforward control of this torque around the elbow might signi cantly contribute to the dysfunction in reaching (36). De cits in the grasp formation during reaching impact as well the reaching time (37). All these complex demands on reaching might increase the within trial variability in reaching seen in individuals with stroke.
To move the hand back to the starting position in the returning phase of the drinking task should theoretically be less challenging, however, up to 8 trials were needed to reach good performance stability in both investigated groups. One possible explanation for this nding could be that the movements in this phase did not require direct visual feedback and that the participants might have corrected the end position of the hand in some trials. To overcome this potential problem, a more standardized end of the task could be used.
The relative time to peak velocity, designating acceleration and deceleration time in reaching, showed also higher variability with 6 or more trials required to reach good performance stability in both groups.
Higher variability, characterized by lower effect sizes of discriminative validity, was also observed for this variable during the drinking task in persons with stroke in a previous study (17). This suggests that this point in time when the peak velocity is reached may vary between trials both in persons with stroke and in those without disability.
Interestingly, in the non-disabled group more trials were needed for NMU (3 to 9 and more) and inter-joint coordination (4 trials) than in individuals with stroke (2-3 trials). The main reason for that was most likely the inherent properties of the variables themselves. In both metrics, the between-subjects' variation was extremely low compared to participants with stroke (see Table 3). Further, the performance of nondisabled participants was also close to the extreme possible value of the metrics (ceiling or oor effect).
These aspects need to be considered when interpreting the reported ICC values for these variables in the non-disabled group.
Good movement performance stability was reached after 2 trials for all joint angles and trunk displacement metrics ( Fig. 1 and Table 4). This nding con rms that movement variability of the joints and segments of the body is relatively stable when repeatedly performing a well-known task (16), such as drinking from a glass, in a self-paced comfortable speed. This result is in line with previous research in non-disabled persons showing high level of automaticity of movement execution of well-learned tasks (16), and also in persons late after stroke where compensatory movement strategies have shown to be more xed (38,39).

Systematic trend over a set of trials
In the non-disabled individuals, small but signi cant trends towards improvement were demonstrated in some temporal variables (for total movement time and for some of the movement phases) when the last three trials were compared to the rst three. These trends might be caused by the learning effect. The improvements were, however, small and can therefore be considered to be of less clinical relevance.
In the stroke group, no signi cant trends over multiple trials were found, but even here small trends could be observed visually in some variables, e.g. increased trunk displacement in later trials (Fig. 2). Not nding signi cant trends in stroke data could be caused by the low power due to the small group size (n = 8), and larger studies in stroke populations are therefore warranted.
We expected to nd signs of muscular fatigue in terms of declining trends in the stroke group over the set of trials, but this assumption was not supported in the results. Interestingly, from an intervention study it was reported that participants in post stroke training could conduct up to 300 repetitions (3 tasks x 100 reps)/occasion, within one hour) without experiencing increased fatigue (40). The risk of fatigue in uencing motor performance after stroke has, however, been highlighted in several previous studies (12,20,22), and a planned rest in between trials has been recommended.
In the current study, the participants took a short break of about 5 seconds between each trial.

Test-retest reliability in a subsample of individuals with stroke
In the current study, good to excellent test-retest reliability with a mean of 2 to 3 trials was demonstrated for most of the kinematic variables in the individuals with stroke performing the drinking task at 4 different occasions. However, for two end-point variables (the peak velocity and the time to peak velocity), the reliability remained poor or moderate even after 9 trials. Our ndings agree with previous research (19,20), even though there are some methodological differences.
In a study with participants with stroke (tested at two occasions, few days apart) good to excellent testretest reliability were found for movement time, peak velocity and trunk displacement in different reachto-grasp tasks (different object sizes and at self-selected and fast speeds) (19). Interestingly, for nondisabled controls only moderate to good reliability was demonstrated (19). The authors proposed that the lower consistency observed in non-disabled individuals might be caused by an exploratory behavior among controls trying to nd the most optimal solutions for movement execution within the existing task constraints (41). Individuals with hemiparesis after stroke often move with behavioral compensation and this altered movement performance has been reported to be less variable (11,38,42). From a theoretical dynamic system's perspective, the underlying mechanisms for these more xed movement patterns developed over time in people with stroke might explain the low observed variations (39).
Test-retest reliability of kinematic variables obtained during a pointing task, using a mean of 2 trials in persons late after stroke, showed varying ICC values (20). Good reliability (ICC > 0.75) was reported for shoulder exion and elbow extension, moderate reliability for peak velocity, shoulder abduction and interjoint coordination, while the ICC values for movement time, time to peak velocity and number of velocity peaks were low (20). In contrast to the Wagner et al. (20), our results showed good reliability for movement time (except for the returning phase) and NMU, while the time to peak velocity showed low reliability similarly to the abovementioned study. Plausible explanations to these inconsistent results might be the differences in tasks and that the participants in the Wagner  week in the current study, which might have in uenced the results.

Strengths and limitations
In the current study a wide range of well-established kinematic variables, covering temporal, end-point, angular and displacement kinematics, were evaluated, which is a strength of the study. The results regarding non-disabled people were based on a relatively large sample (n = 44), although the results from stroke participants need to be interpreted with caution due to the small sample size (n = 8). However, the kinematic variables analyzed in the current study in stroke participants showed a consistent pattern in line with existing research (11). This implies that 3-5 trials per test occasion might be used as a rough guide for self-paced functional everyday reach-to-grasp tasks both in non-disabled people and in individuals with stroke.
As also experienced in the current study, not all trials might be successful during the data capture due to various reasons including obscured markers and data gaps. This might be particularly relevant for individuals with stroke where the altered movement patterns might cause obscured markers resulting in data gaps. This further suggests that even when a good performance stability might be reached with 2 to 3 trials, few extra trials are needed to ensure su cient number of successful trials.
The results of the current study are only applicable for the kinematic motion capture systems using multiple optoelectronic cameras. The results seem, however, to be similar even when the kinematics are collected by other systems, such as Kinect camera or inertial sensors (32,33). This is promising, taking the constant push from users (clinicians, researchers and patients) to make movement analysis more readily available with systems that can operate outside the lab.

Conclusions
This is the rst study that determines the number of trials needed for good performance stability and testretest reliability for kinematic variables during a reach-to-grasp task in persons with and without upper extremity impairments. The ndings in this study demonstrated that only 2-3 trials were needed to reach good within session performance stability for most of the kinematic variables of the drinking task, both in non-disabled persons and in a sample of individuals with chronic stroke. The small trends towards shorter movement times between rst and last trials that were observed in non-disabled individuals were not detected in the participants with stroke. Good to excellent test-retest reliability (comparing 4 occasions) was also reached for many of the retrieved kinematic variables when a mean of 2 to 3 trials were used in a subgroup of individuals with stroke.
These results imply that a recommendation for future studies to collect at least 3 trials of each tested condition is well founded and applicable for most of the kinematics.
However, there are few exceptions, and in these cases a larger number of trials is warranted. The results are primarily applicable for the drinking task, but partly also to other similar reach-to-grasp tasks.

Declarations
Ethical approval and consent to participate The ethical approval was provided by the Regional Ethical Review Board, Gothenburg, Sweden (318-04) and the Swedish Ethical Review Authority (1074-18, 1075-18). The Helsinki declaration was followed and an oral and written informed consent was obtained from all participants prior data collection.