Invited Commentary

There is an increasing interest in the use of robotic-assisted therapy to facilitate and augment upper-extremity movement for people with hemiparesis. This focus is an attempt either to improve patients' rehabilitation outcomes beyond our current capabilities or, as Kutner et al¹ suggest, to decrease the therapist time demands necessary for the delivery of repetitive task practice. The robotic assistive devices have been used in accordance with current neuroscience literature in animals and motor control literature in humans. They take advantage of recent improvements in robotic design, the development of haptic interfaces, and the use of virtual reality simulations, interfaced with the robots. Most of the robots have been designed to train the shoulder and elbow, supporting the limb as needed, with the limb moved either passively^2,3 or requiring active movement.^4–7 Far fewer robots have been designed to train the arm and wrist⁷ or arm and hand together.^8–12 Some of the robots emphasize impairment-based practice,^3,4 whereas other robots focus on more task-based practice,^7,9,10 with the complexity of sensory feedback ranging from simple⁴ to complex, 2- and 3-dimensional, interactive simulations.^9–14

Two reviews found that robot-assisted therapy showed potential to improve upper-extremity function^15,16 and improve strength.¹⁶ However, neither review could confirm evidence for improvement in activities of daily living, which may reflect an inability of activities of daily living scales to accurately demonstrate changes in paretic limb function.¹⁵

In this study, Kutner et al showed that 30 hours of therapist-supervised repetitive task practice combined with 30 hours of robotic-assisted therapy resulted in similar reports of improvement in hand function, as measured by the patients' responses to 5 questions on the Stroke Impact Scale (ie, carrying heavy objects, turning a knob, opening a can/jar, tying a shoelace, and picking up a dime), as well as improvement in the measure of overall stroke recovery. The robotic device used in this experiment was the Hand Mentor.

The Hand Mentor has preprogrammed activities that provide continuous stretch of preset durations to the finger flexors (termed “anti-spasticity program”), a wrist flexion or extension strengthening protocol, and a muscle recruitment protocol, where the patient receives feedback regarding the intensity of the electromyographic signal related to wrist and finger movement. As this is an impairment-based system, it appears that, in addition to comparing the use of the robot as a therapeutic adjunct to task practice, the authors also were comparing particular therapeutic principles, specifically task-based training with a combination of task-based training and impairment-level interventions. The authors' findings of similar improvements in reported hand function and stroke recovery raise several interesting questions. What is the role of impairment-level interventions versus functional training? Is the combination important, and if it is important, what is the ideal dosing for each component? Is the total intensity or amount of practice more important than the difference in approach?

It is worthy to note that assessment of hand function improved over time, with no significant difference between the 2 intervention groups; however, the mean of the combined therapy substantially exceeded the minimal clinically important difference in the hand function score, as determined in a recent study by Lin et al.¹⁷ Kutner et al acknowledge that the exact mechanism responsible for improved rating of hand function is not known. They believe it is likely that an increase in the active range of motion of the wrist and fingers and a possible reduction in spasticity of the hemiparetic limb may have improved the patients' perception of hand function, thus suggesting an essential role for impairment-based interventions. Given the currently strong interest in task-based therapy, the possible influence of impairment-level interventions is an important point to be remembered.

Another essential question is whether robot-assisted therapy would be better used for impairment-level interventions or for task-based training? An economic argument might favor their use for task-based training. The literature regarding the conditions for success of task-based training indicates the need for highly repetitive, long-duration interventions.¹⁸ This type of intervention is extremely labor intensive and time consuming for clinicians, well beyond what the current health care environment can or will provide, thereby arguing for much-needed technological assistance for these interventions.

The participants in this study were trained within 3 to 9 months poststroke. Studies on patients with subacute stroke are important because of the increased potential for improvement at this point in time. However, a concern during this time frame is how to separate changes in function resulting from the intervention from changes due to spontaneous recovery. Double baseline testing, separated by a period of time, often is used to control for this issue.

Although quantitatively the outcomes were similar between the 2 groups in the measure of hand function, clinicians would have liked to be able to discriminate between recovery based upon compensation and recovery based upon true improvement in motor abilities of the hemiparetic limb. This is an especially important issue in regard to hand function and is not possible with information provided through the Stroke Impact Scale. Patients consistently report compensatory use of the uninvolved hand for activities of daily living when they are even slightly unsure about the reliability of their hemiparetic hand function. Additionally, it is impossible to know whether the increased score was due to proximal changes in the shoulder or elbow or to distal changes in the wrist or hand. Although Stroke Impact Scale hand function subscale scores show a moderate correlation to tests of upper-extremity function and have been found to be responsive to changes in hand function,¹⁹ it is not possible to obtain important information from a quality-of-life measure such as the Stroke Impact Scale. Kinematic and kinetic analyses and data from timed measures of hand and arm skill, such as the Jebsen Test of Hand Function²⁰ and the Wolf Motor Function Test,²¹ would have greatly contributed to our understanding of the sensorimotor changes found in each group in this study and would have provided a context for understanding the implications for therapeutic decisions and interventions. Even reports of motoric changes in individual patients are helpful for clinical decision making.

Development of this type of complex technology and its use as an adjunct to existing therapies is intended to improve therapeutic outcomes; however, progress is costly and, therefore, slow. Development of robots to assist hand function is particularly challenging. Inconsistent use of the same outcome tests and measures and kinematic analyses hinders interpretation of results and impedes progress of the development and incorporation of these technologies.

• © 2010 American Physical Therapy Association