Movement recovery after stroke is a long process requiring intensive rehabilitation. As an individual recovers, the setting of their care changes and the level of supervision decreases, from supervised clinic, to supervised home, to unsupervised home rehabilitation. Robotics and sensor-based technologies rehabilitation technologies (RTs) have been developed to support rehabilitation through these stages, but it is still unclear when and how they can be most useful. This dissertation leveraged systematic literature review, qualitative data analysis, and direct measurements from RTs themselves in the form of usage analytics to gain insight into these issues and improve RT design. In Part 1 of the dissertation, use of RTs was studied in the three settings. To understand uptake in the supervised clinical setting, a systematic review of the existing literature on rehabilitation technology was used to generate a list of 17 constructs that influence uptake. Three occupational therapists (OTs) and two physical therapists (PTs) employed at a major, rehabilitation hospital that encourages the use of technology wrote vignettes from a written prompt describing their RT use decisions during treatment sessions with nine patients. Vignettes were coded using deductive qualitative analysis from the 17 constructs. In the clinical setting early in rehabilitation, therapists rarely chose to use RT, characterizing candidate RT as having a relative disadvantage compared to conventional treatment in the clinic due to lack of relevance to functional training, the time required for training and setup, and poor adaptability to patient-specific attributes, such as cognitive limitations. RT was typically only used when specific devices provided a desirable function that conventional treatment could not.
In contrast, RT was seen as having a relative advantage during supervised use in the home setting. A Sensor Enhanced Activity Management (SEAM) system was developed to combine home exercise program (HEP) management software with a movement sensor for monitoring and motivating HEP adherence. Three therapists used the system in their regular practice during the first six months of the COVID-19 pandemic. Patients were active for a mean of 40% (26% SD) of prescribed days and completed a mean of 25% (25% SD) of prescribed exercises. The therapists reported that remote monitoring and the use of a physical movement sensor was motivating to their patients and increased adherence, highlighting a perceived advantage of RT in the home setting.
Given RT’s potential in the home setting to motivate use, can changes in the design of RT modulate uptake and further increase its use? Motor learning research suggests that a key to motivation is optimizing challenge. This hypothesis was tested using long-term, self-determined exercise patterns of a large number of individuals (N = 2,581) engaging in home rehabilitation with a sensorized gaming exercise system (SGES) without formal supervision. The SGES is comprised of two puck-like sensors and a library of 40 gamified exercises for the hands, arms, trunk, and legs that are designed for people recovering from a stroke. Appropriate challenge level and regular initiation of exercise sessions were found to maximize perseverance, while experiences of low challenge in the first week were associated with the lowest levels of overall perseverance.
In summary, Part 1 of the dissertation found that RT has a relative advantage in the home versus clinical setting if it can improve motivation. Further, a key factor of RT design – the challenge level it presents – was demonstrated to modulate perseverance. Part 2 of the dissertation then focused on measuring the actual effects of RT uptake in the home setting. First, a systematic review was conducted and the SGES was evaluated based on recommended design features. The SGES can be considered near optimal, except it does not have a recommended feature that allows a therapist to monitor and communicate with the user. Next, improved clinical outcomes resulting from use of the SGES compared to conventional home therapy were demonstrated based on analysis of results from a single-blind, randomized controlled trial with 27 participants in the subacute phase of stroke. Adherence in the context of the clinical trial was shown to be superior to adherence in an unsupervised context, suggesting that the missing design feature of therapist presence affected adherence. Finally, it was demonstrated that home RT has the potential to measure its own effect on the user, because it provides data that can be used to accurately estimate users impairment level using nonlinear modeling techniques. This opens the door for self-assessing RTs for the home setting.
In Part 3 of the dissertation, refinements to the design of the SEAM and SGES RTs for home exercise were implemented and evaluated. Changes to the SEAM system focused on automating the connections to the sensor via Bluetooth and improving the repetition counting algorithm. In a subsequent trial of the system, 63% (39% SD) of the exercises that patients completed in the system generated sensor data compared to 22.1% (29.7% SD) of the completed exercises from the first trial, a three-fold increase in the uptake of the sensor component. A suite of changes to the SGES studied here aimed to ramp up the difficulty more quickly for less impaired users and give an option to manually lower the difficulty for more impaired users. However, examining a second set of user data (N = 1,251) 2.5 years after these changes were implemented showed a decrease in overall levels of perseverance relative to the previous versions of the system in the number of repetitions performed (median 2,047 vs 1,219 repetitions), time spent exercising (median 52 vs 48 minutes), and the number of days (median 6 vs 5 days) the system was used. Though the intent of the changes was to have users experience more challenging exercises more quickly, the actual result was that more users achieved 100% success during their first week of use, an experience which corresponds to low rates of perseverance in the system. The reasons for this unintended result are discussed, but, overall, this study illustrates at large scale how RT software design changes modulate user behavior in complex ways. Further, the self-monitoring capability of RT is proposed as a powerful new tool to guide RT design for the home rehabilitation environment, where measures of adherence have been historically difficult to obtain.
