This paper presents the clinical feasibility and outcomes of a RCT conducted on subjects with subacute stroke evaluating the effect of robot-assisted neurocognitive therapy of hand function, and in particular, if therapy with the haptic device could lead to an equivalent and lasting sensorimotor recovery compared to dose-matched conventional neurocognitive therapy. In contrast to most robot-assisted rehabilitation trials, which placed a strong focus on movement training, our approach takes full advantage of the haptic rendering abilities of the robot, and proposes a therapy program adapted to these capabilities. We could show that this approach is well accepted and could be integrated in the daily schedule of inpatients in the subacute stage after stroke. Most participants found the program motivating, comfortable, and could perceive concrete improvements in their health status after the end of the treatment.
Equivalent reduction in upper limb motor impairment
Traditionally, most RCTs have aimed to prove that robot-assisted therapy per se could increase upper limb recovery with respect to conventional therapy (e.g., by increasing therapy intensity, subject engagement, or by providing exercises targeting specific motor impairments). However, large clinical studies on arm rehabilitation with subjects with chronic stroke, aiming to demonstrate the superiority of robot-assisted therapy, were rarely successful, or only observed small, non-clinically meaningful differences [2, 3]. Similarly, other studies focusing solely on robot-assisted rehabilitation of hand function in chronic [48–51] or subacute stroke [12, 52, 53] were not able to show statistically significant differences between robot and control therapy groups, or reported minor differences in secondary outcome measures [10, 54, 55]. The present RCT directly investigated equivalence in motor impairment reduction between a robot-assisted and a conventional therapy group focusing on the training of the upper limb, and in particular the hand. For this purpose, the therapy dose (i.e., total number of task repetitions and therapy time) as well as the therapy intensity (i.e., task repetitions per time unit) were precisely matched between groups.
The results of the equivalence test comparing the evolution in FMA-UE demonstrate that, for our specific intervention, the motor recovery in the robot-assisted group is non-inferior with respect to the control group. In general, it is not surprising to observe little to no difference between conventional and robot-assisted therapy in the context of studies where therapy dose and the therapy exercises/movements are designed to be similar, qualitatively and/or quantitatively, between groups. In this context, the haptic device primarily supports the therapist, providing additional motivation for the subjects to train, and delivering objective readouts (e.g., based on task performance, or kinematic and kinetic data) that can be used for monitoring, difficulty adaptation, or research purposes [29]. Nevertheless, the fact that a session of conventional therapy could be replaced without affecting the overall rehabilitation outcome opens promising avenues for further developing robot-assisted therapy programs.
Neurocognitive hand rehabilitation led to improvements on motor, sensory and cognitive scales
Secondary outcome measures further support the equivalence analysis. After four weeks of treatment (T1), in addition to motor deficits, also sensory and cognitive deficits were concurrently reduced in both groups, with improvements in all the secondary clinical scales (i.e., proximal and distal arm impairment, functional ability, somatosensation, executive functions and cognitive control). The decrease in upper limb impairment (FMA-UE, RG + 7.14 pts, CG + 6.85 pts) was clinically meaningful in both groups, and favorably compares to other work focusing on robot-assisted hand rehabilitation in subacute stroke, where changes between 3.0 to 5.3 FMA-UE points were typically reported [9]. Improvements were retained over time up to 7 months after the end of treatment. No significant differences were found between the groups in terms of changes with respect to baseline for all outcome measures. Only the BBT showed a significant effect of time, with additional increases in changes with respect to baseline after completion of the intervention (i.e, above 98% at T3). FMA-WH also showed steady improvements over time after the end of the intervention (i.e. after T1), but these were not significant after Bonferroni correction. These further increases in BBT and FMA-WH suggest improvements in unilateral gross manual dexterity, which represents an essential element in the interaction with objects. This supports the approach of, whenever possible, focusing therapy on hand function training rather than proximal arm segments only, as distal training may promote impairment reduction in the entire arm [13, 14, 56].
Only minor improvements were observed in both groups over time in cognitive functions (i.e., FAB, MMSE, Albert Test), somatosensory function (i.e., EmNSA), and muscle tone control (i.e., MAS). These changes were small mostly due to the saturation of these scales in a mildly/moderately impaired population, and did not show significant changes between the groups and over time following T1. A decrease in MAS was observed in the CG, but not in the RG where a small, clinically non-relevant increase was observed. This is partly in line with a recent review [1], which analyzed changes in MAS of the paretic arm in 13 dose-matched RCTs and found negative effects on muscle tone reduction (i.e., increase in MAS) following robot-assisted therapy and a significant difference in favor of the respective control groups. This could possibly be caused by higher forces/muscle recruitment involved in robot-assisted exercises, but was not monitored in the present study. Also, it is debatable whether a minor, temporary, increase in muscle tone would negatively affect functional recovery in subjects with stroke [17, 57]. In the present study, the increase in MAS disappeared in follow-up assessments, and the different behavior of the two groups could also be explained by slightly higher baseline MAS in the CG. Additional studies are necessary to investigate how muscle tone evolves depending on subject conditions (e.g., lesion type), therapy type and intensity.
Study Limitations
The participants involved in both groups were mostly mildly or moderately impaired (initial FMA > 29) [43]. This led to ceiling/floor effects in some of the clinical assessments, which might have masked some of the intervention effects. This was, however, not imposed by our study design, subject screening or inclusion criteria, as the feasibility of the proposed robot-assisted therapy approach was also demonstrated in more severely impaired outpatients in the chronic stage after stroke [28]. While the robot-assisted therapy program could be well integrated into a subacute rehabilitation program to complement the existing therapy, only subjects with mild to moderate cognitive impairment were eligible to participate, as the intense therapy program challenged some subject’s cognitive abilities. As additional possible confounder, all participants received additional conventional therapy sessions as part of their standard inpatient therapy program in parallel to the intervention, which could not be entirely substituted for ethical concerns. Nevertheless, we did reach a dose of 15 × 45-minute sessions over 4 weeks, which is comparable to other clinical trials or pilot studies on robot-assisted rehabilitation of hand function [51, 53, 54]. Furthermore, the results of this study are limited by the rather small sample size and should be interpreted with respect to the provided therapy and dose level. Finally, as is the case for any clinical trial at the subacute stage post stroke and of comparable sample size, the contribution of spontaneous recovery cannot be disentangled from intervention-induced recovery.
Potential of neurocognitive robot-assisted rehabilitation of hand function
The therapy intensity delivered in this study typically exceeded the amount of movement practice reported in the literature for conventional physio- or occupational therapy sessions (1.45 rep/min vs 0.92 rep/min) [58]. Still, compared to the knowledge gained from animal studies and to recent high-dose clinical studies [59–61], this intensity might not be sufficient. It is important to note, however, that one “repetition” using the neurocognitive approach is not directly comparable to, e.g., reaching movements as typically reported in the literature. One repetition corresponds to one complex task (e.g., sponge identification) involving several actual movements, as well as sensory processing and cognition, demanding time, effort and concentration. An open question is whether delivering high (and potentially even higher) intensity of conventional therapy would be feasible in daily practice over several weeks, outside of a research study.
By establishing the non-inferiority in impairment reduction in the two groups at a clinically-applicable dose, the proposed robot-assisted therapy of hand function including sensorimotor and cognitive training opens new research avenues, as robots could offer a viable approach to further increase therapy dose outside of therapy programs [62, 63] with minimal added burden on clinical staff. This could, for example, be achieved through semi-supervised therapy during inpatients’ spare time [64], or even directly at home, as proposed in several promising pilot studies with passive devices [65, 66]. In that sense, the results of this study demonstrating that neurocognitive robot-assisted therapy is also safe and well-accepted are a positive and necessary first step.
To reach the goal of minimally supervised robot-assisted rehabilitation, special attention should be devoted to the evaluation of usability and acceptance of rehabilitation devices, and in that sense, simple end-effector devices, such as the ReHapticKnob device used in this study, may be advantageous over upper limb exoskeletons often requiring long setup time and adjustments [9]. To meet user expectations and improve technology acceptance with respect to the current setup, more attention should be devoted to the design of ergonomic handles and to the adaptation of the difficulty levels of the exercises. Embedded clinical “intelligence” building on online robotic assessments, or performance metrics extracted from therapy sessions, should be further developed to provide means of accurately monitoring subjects’ ability level and the evolution of their performance during treatment (or even after discharge), possibly adapting exercise difficulty autonomously to constantly challenge the user at an appropriate level [29]. Overall, such technology could help to increase the therapy dose subjects with stroke receive at the different stages of their rehabilitation, offer alternative solutions to enable earlier discharge of the subjects from the clinics, and provide continued and individually modulated therapy in the home environment after discharge.