4.1. Feasibility of AGS
The primary aim of this study was to examine the feasibility of the AGS exercise. All ten patients regardless of their diagnosis or level of ambulation were able to train with AGS. The findings indicate that AGS training is feasible and safe in a long-term application in clinical practice. Due to the characteristics of the study site, which was an outpatient training center, only outpatients in the chronic phase of their disease were included in the study. Given that the range of impairment was very high (FAC levels ranged from 0 to 5), it can be assumed that AGS would also be feasible with patients in a (sub-)acute phase, provided that they are not cognitively impaired.
These results were also confirmed by the positive answers to the TAM questionnaire. Evidence in the literature also indicates that patient-cooperative training with visual feedback is well accepted by patients, improves their active participation, and means that they are able to perform these kind of tasks [
9,
37].
4.2. Adaptive Gait Support as a Training Tool
This study also aimed to investigate the characteristics of AGS as a training tool. The difference between the robotic support determined automatically by the AGS and the support manually determined by the therapist during conventional Lokomat training was also examined. Additionally, the behavior of the support throughout the gait cycle was examined.
In almost every gait phase, all patients reached significantly lower levels of robotic support during AGS training than during conventional Lokomat training, where the support was set manually by the therapist. Thus, regardless of the diagnosis or level of impairment of the patient, the adaptive controller set lower robotic support levels during AGS training than the therapists set. It has to be noted, however, that during conventional Lokomat training the therapist was only asked to set the robotic support to a level which he/she thought was best for the patient and not to lower the support to a minimum, while during AGS, the controller determined the most suitable robotic support for each patient, and for each gait subphase. Despite recommending that the patient is challenged as much as possible, it is not unusual for therapists to maintain a high robotic support (“guidance force”) during clinical practice. The therapists often prefer to be on the safe side and prevent the Lokomat from stopping (if the robotic support is set too low, kinematic deviations may trigger the safety mechanisms). In addition, due to the missing physical interaction between patient and therapist, the therapist can have difficulty determining how active the patient in the machine is. One reason for the high reduction in guidance force achieved in this study could be that AGS allows a higher support in the gait phases where it is more needed, while a low support can be reached in others. When the therapists manually set the guidance force in the standard Lokomat controller, they can only lower it as far as the weakest phases of the patient’s gait cycle allow. By calculating the kinematic deviation between the actual and reference foot trajectory in 30 different windows of the gait cycle, the AGS controller supports the therapist by finding the performance limit of the patient in every phase of the gait cycle, thereby providing targeted assistance throughout the gait cycle. Moreover, as the AGS software adapts the robotic support at every session, it can address the changing needs and improvement of patients, who are expected to improve during the course of rehabilitation, thereby requiring less and less support from the machine.
In the literature, it has been shown that robotic devices may lead to slacking (i.e., the patient’s effort is decreased during training because the robotic device provides too much assistance) [
10] which could potentially lead to a decreased recovery. During training with AGS, the robotic support can be lowered automatically, which may prevent the above-mentioned issues. However, further research is needed to test whether AGS can prevent slacking, as well as to examine whether AGS is superior compared to other robot-aided gait trainers and controllers in terms of optimal challenge of the patients.
When investigating the applied robotic support values on the Lokomat reported in other studies, we see that the minimum support values that could be reached in most studies were generally higher compared to the support levels reached in this study using AGS. Additionally, it was stated that attempting to lower the support to a minimum is especially challenging [
38,
39,
40]. An explanation for this discrepancy could be the different patient population or the fact that the other studies could not appropriately target the different gait phases, maintaining a constant level of support throughout the gait cycle. However, Chang et al. and Mayr et al. lowered the support even more than in the current study but only included unilateral stroke patients, and it is not clear how many patients reached the minimal value [
41,
42].
4.3. Adaptive Gait Support as an Assessment Tool
A further aim of the study was to investigate the performance of AGS as an assessment tool by comparing its data to conventional walking tests and isometric force assessments and calculating the reliability of the assessment.
The analysis of the variation in walking ability explained by the AGS data resulted in a significant relationship between the measures of clinically established walking tests (10MWT and TUG) and AGS data (represented by robotic support at initial swing) focusing on the overall patient effect. The choice of the initial swing phase as a predictor of clinically established walking tests is in line with what is reported in the literature about this subphase—namely that the initial swing phase is a critical phase of the gait cycle [
43]. The performance at the initial swing phase is particularly influenced by an appropriate push-off, which is one of the main determinants of gait speed and is very often impaired in neurological patients. During ankle push-off, most of the power during the gait cycle is generated [
33].
Even though the results of this study showed no linear relationship between the isometric force (L-FORCE) and AGS data, we could hypothesize that the two measures reflect two separate components of walking ability: motor control (i.e., the ability to modulate the force) and the ability to generate high forces, as also suggested by Maggioni et al. [
22]. The two measurements together can subdivide the patients into clusters (e.g., showing high or low isometric force or showing good or poor motor control). A clear separation between ambulatory patients and non-ambulatory patients, based on the maximum force the patients were able to achieve, could be made (see
Figure 7). However, within the non-ambulatory patients, some reached low knee stiffness values in the AGS: this could be an index of a partially regained motor control with a persistent weakness. Research suggests that in order to regain walking ability, not only force [
44] but also control is needed [
5]. In Maggioni et al. [
22], AGS and L-FORCE measures together increased the explained variance in a model with timed walking tests, suggesting that not only motor control (as measured by AGS) but also force (as measured by L-FORCE) was needed to explain the variance in walking ability expressed in a clinical test such as the 10MWT. Therefore, these two measures together could be an indicator of the patient’s progress from a non-ambulatory to an ambulatory stage and could be an additional tool for the therapist to evaluate how far the patient is from walking again. This possibility would be particularly relevant for patients with severe walking impairments, who have few other options for objective instrumented assessments of lower limb impairment and rely only on ordinal, coarse scores, such as the FAC.
When observing the progress of robotic support within the same patient over the 8-week training period, we did not observe trends in decreasing robotic support: this may be because patients were chronic, so little improvement in walking ability was expected, and therefore nothing could be captured by AGS. We also noticed a high variability between sessions. This may be for several reasons: AGS requires the patient to follow a trajectory displayed on the screen, a task that requires concentration and proper visuomotor coordination; therefore, distractions may influence the results. Moreover, patients had fluctuations in their health status due to symptoms such as inflammation and spasticity, also resulting in changes in performance during training.
The reliability analysis resulted in a good relative (using classification suggested in the literature [
45]) but poor absolute reliability. This absolute reliability of the assessment should be improved before AGS can be used as an assessment in practice. The data which were considered for this analysis were one week apart for some patients, which may have led to increased variability due to patients’ differing daily health condition. In a future study, the reliability should be tested in sessions within one day or on two consecutive days, which may lead to an improved absolute reliability.
The AGS exercise can be integrated into the training procedure. To our knowledge, no commercially available robot-aided walking assessment exists, which is feasible in everyday practice without requiring additional time above that of the therapy.
Since AGS can be used with non-ambulatory patients in the Lokomat, before they can even perform a walking test (e.g., 10MWT), AGS may be able to assess the progress from the non-ambulatory to the ambulatory phase. An assessment for non-ambulatory patients is highly needed because, in terms of functional improvement, there are no feasible and valid performance-based measures of how close non-ambulatory patients are to walking again. Most clinical assessments are too crude to reveal functional changes [
46] and lack the sensitivity to detect small changes in the rehabilitation process. A quantitative and responsive walking assessment for non-ambulatory patients might help to keep the patients motivated, as well as support the therapists in making data-driven decisions on the training program. However, the concurrent validity of such an assessment cannot be proven against established walking tests such as those used in this study (10MWT, TUG), as non-ambulatory patients would not be able to perform them.
4.4. Future Directions
Including participants in the (sub-)acute stages would be reasonable to detect larger changes in clinical scores and therefore a potentially significant effect within the patient. A clinical assessment must demonstrate a high responsiveness in order to be useful in clinical practice, and this aspect needs to be further improved and studied in future research on AGS.
The AGS assessment could be used to measure the asymmetry of the paretic and non-paretic leg by calculating the ratio of the legs. This ratio could be used as a helpful feature in addition to the L-FORCE data, which has been shown to be able to detect significant differences between the paretic and non-paretic leg [
47]. Thus, the symmetry of the legs in terms of strength and control could be assessed.
No previous investigations have assessed the application of AGS in practice in a longitudinal study. This is the first study in this domain. These results support additional large-scale studies, where grouping the patients by diagnosis could provide more insights on the recovery trajectories of specific diseases. Further investigations should focus on the training effect of AGS and on improving the psychometric properties of AGS as an assessment tool, to enable it to capture the rehabilitation progress of non-ambulatory patients and therefore be used as a valid and reliable assessment tool. In principle, AGS as a training tool demonstrates beneficial features, such as requiring the patient to contribute actively to the conducted movements by lowering the robotic support in phases where less support was needed. These features could possibly contribute to more intensive and effective training. Additional studies are needed to provide evidence for this.
4.5. Limitations
This pilot trial has several limitations that restrict the strength of its conclusions. First, the sample size was small, which made it difficult for some parts of the analysis to achieve significant results. In addition, performing subgroup analysis based on patients’ diagnosis was not possible with this sample size. The diverse patient characteristics and their possibly different responses to Lokomat therapy could potentially have influenced the results.
Secondly, only chronic patients were included in the study. Therefore, no big changes between the assessments of walking ability were seen, which made it difficult to observe changes in robotic support during the course of the study.
Another limitation was the fact that AGS training sessions were not controlled for BWS. The therapist tried to maintain the same level of BWS throughout the study period, but it could not be guaranteed that the same levels were applied in all sessions. This may have influenced the robotic support during AGS training which could be a reason for the high variability in the AGS data.
Lastly, the AGS controller was only applied to one leg at a time, potentially leading to compensatory strategies to cope with a challenge on a single leg. While adapting the robotic support for both legs at the same time might prove too challenging for some patients, future analyses should address the effect of this choice.