It was feasible and well-tolerated to incorporate an additional three hours per week of gait training therapy into all participants’ schedules over a large portion of their inpatient stay. On average, study participants received 11.3 supplemental therapy sessions (12.8 CGT, 9.8 Lokomat®). It has been noted that other aspects of each individual’s plan of care and stage of recovery can be an obstacle to supplemental therapy [19] but with a clear focus and charge to incorporate additional gait training, finding three extra hours per week proved to be practically achievable during early post stroke inpatient stay time for this small cohort. This was completed during “normal care hours” and did not require night or weekend time to achieve. It is worth noting that current mandates are that IRF stroke patients receive at least three hours of therapy per day, five days a week during their inpatient stay. This therapy is divided between physical, occupational and speech therapies – in our institution, at about 1.5hours, 1 hour and 0.5 hours, respectively. The physical therapy is then split, as needed, to address functional limitations and goals as appropriate for each patient. As such, three hours of additional gait training per week represents a significant increase over a patient’s existing therapy regimen.

Enrolled participants tolerated additional exercise well. The cardiorespiratory inclusion criteria for study participants biased us to expect this. There was a slight reduction in therapists' perceived exertion in the entire cohort over the course of the study. Neither individual treatment group analysis showed statistical significance given the small group size, the relatively small change (0.7 for each group) and the relatively high variability (1.1 Lokomat®, 2.4 CGT) of that change within each group. The Lokomat® group showed a medium effect size (0.62).

Receiving supplemental therapy of either form led to marked improvement across many of the study measures (p<0.005 in 9 of 20 comparisons, Table 2). To further support this, we compared study data to data from our reference group who did not receive supplemental training. This large group of 415 diverse patients had an average age of 53.1 years (standard deviation, 5.1 years), were 60% male, and 85% had an ischemic stroke (15% hemorraghic). These data suggest a fairly similar group to the current study group – though not a true matched control. They were, overall, slightly younger than the entire study cohort but about the same age as the CGT group (both of whom were about 10 years younger than the Lokomat® group), slightly less %male (60% male vs. 73% for the study group), and slightly greater % ischemic stroke (85% vs. 73% for the study group). This reference group had a higher average intake FIM motor score (31) and a similar average discharge score (51.5) for an average improvement of about 20.5. In contrast, the study participants’ FIM scores improved 24.5 for the Lokomat® group and 26.6 for the CGT group, both representing a meaningful increase over the reference group FIM improvement [20]. As such, our findings support that the supplemental gait training was valuable to improve mobility function at discharge based on FIM outcome measure.

The reference group having higher intake average FIM score can be viewed in two ways. First, having higher function may allow them to achieve more during gait training. Alternatively, the lower the intake score, the more potential room for improvement. The latter idea is consistent with the general finding that lower functioning (non-ambulatory) participants obtained the greatest benefits from electromechanical assist in training [21] and supported by findings such as change in self-selected velocity (SSV) being negatively correlated to SSV at intake [22]. Also, the stroke subtype has commonly been thought to affect functional prognosis where ischemic stroke patients tend not to improve as well during therapy. Recent work suggests that this may not be true – and so perhaps this discrepancy in our reference and research cohorts is not so important [23,24,25]. Some of these factors could be contributing to differences observed in outcomes between our reference and study groups. However, our data clearly indicate that both groups improved considerably during their inpatient rehabilitation stay and, at least in the comparison we were able to make, that improvement was likely, at least in part, due to the supplemental training they received.

One additional benefit of the supplemental therapy was that both groups showed an absence of a decrease in the PROM and an absence of an increase in spasticity (MAS) of the paretic side ankle and knee joints (Table 2, Ankle DF, Knee Ext, Ank MAS, Knee MAS columns were not statistically different (admission vs. discharge) for both groups). Reduction in PROM is known to occur immediately post-stroke [26] due to development of an upper motor neuron syndrome, the resulting spasticity, reduced motor control and activity limitations. Preventing this decline can contribute to lessened functional limitations both immediately and long-term. PROM was found not decreased for the ankle (Lokomat®: mean, µ=+4.4 degs, WSR p=0.77, CGT: µ=+1.3, WSR p=0.64) and for the knee (Lokomat®: µ=+1.3 degs, WSR p=0.25; CGT: µ=− 0.3 degs, p=0.75). Note, a positive value (i.e. +4.4) indicates the PROM increased slightly, so for both groups, ankle PROM and for the Lokomat® group knee PROM actually increased slightly. Spasticity, gauged by MAS, slightly increased (not statistically significant) for ankle in both groups and knee in the CGT group. Increases were modest, between 0.2 and 0.4, on average per group.

Outcome variables linked to functional gains showed the majority of the improvements in both groups. FAC, FIM, 5xSTS, 2MWT, and 10MWT showed large effect sizes (>0.8) in both groups – as well as a trend towards, though not statistically significant, differential effect between groups. FAC and FIM showed average improvements of greater than 1.25 and almost 25, respectively. The average FAC at admission was between 1 and 2, indicating participants required continuous support for weight bearing/balance or intermittent support with balance/coordination during walking. The average score at discharge was just below 3, indicating the requirement of physical support for walking was mostly eliminated. As noted previously, the 5×STS times also improved considerably. There was a notable differential in improvement in STS time between the groups. The Lokomat® group improved by just over 50% (45s at admission vs. 22s at discharge) whereas the CGT group improved by just under 30% (29s at admission to 21s at discharge). We wish to highlight that 9 participants could not complete the STS task at admission and 5 of these could at discharge (see note a). Though these data did not contribute to above reported % changes and statistical analyses, their functional gains were just as important, if not more so, than being able to stand more quickly – as was reported for the rest of the cohort. The 2MWT also showed similar marked improvements. The entire cohort, on average, was able to quadruple walking distance from just under 10m at admission to slightly over 40m at discharge. The difference is even more striking when segregated by intervention. The Lokomat® group improved nearly ten-fold, from 3m at admission to 29m at discharge and the CGT group more than tripled (16m at admission to 56m at discharge). A similar picture emerged with the 10MWT. Though the overall group change in walking velocity was 0.24m/s, the Lokomat® group went from 0.02m/s to 0.26m/s and the CGT group went from 0.17m/s to 0.40m/s. Improvements in both groups are substantial and clinically meaningful[27]. The Lokomat® group improvement by 50% in 5×STS time and near 10-fold for 2MWT distance and 10MWT velocity are remarkable and prompted post-hoc analysis of differential effect in modality.

Overall, an important point this work helped to establish is that standard of care plus supplemental therapy can lead to many desired improvements and lack of decline, and this can be done in a fairly efficient manner during the inpatient stay without undue burden on therapist (vis-à-vis reduced RPE scores). Additional studies need to be done to further refine the supplemental protocol, including defining optimal intensity and duration, patient demographics as well as delivery mode. The use of robotics can be an important factor in the practical implementation of supplemental therapy as trying to balance therapist time and physical demands with increasing dosage, timing and scheduling during increasingly shorter inpatient stays can be challenging.

Based on the above noted possible differences in effects between treatment groups, we formally explored whether there was a statistical effect of modality. A post-hoc overarching statistical analysis of robotic vs. conventional (CGT) therapy did not show a statistically significant differential effect (MANOVA p=0.21, Pillai Trace). We compared only the interval variables (ankle and knee PROM, 2MWT, 10MWT and 5xSTS) using a single factor MANOVA with post-hoc ANOVAs to compare differences in individual variables. Only these 5 variables were compared since a MANOVA analysis requires interval variables. Though not significant, two of these five variables did show medium to nearly large effect sizes (Cohen’s d of 0.56 for knee PROM and 0.77 for the 5xSTS). Medium to large effect sizes yet not significant p-value (MANOVA) may be due to the relatively small sample size and short duration of the intervention (up to an extra three hours per week for approximately three weeks). Because of the relatively large differences in dose between groups, the regression analysis was also evaluated for FIM, FAC, 10MWT, 2MWT and 5xSTS. Only the FAC was significantly correlated with dose (Table 2, last row). Thus, only for FAC did the covariate have a significant effect. For FAC only, the ANCOVA was computed with number of training sessions as the covariate (Table 2, Note #4). The partial eta squared (h2) value of 0.17 indicates that dose accounted for only 17% of the variance in the FAC score. The adjusted means did grow more different, supporting that the Lokomat® group improved more (mean change of 1.27 vs. 1.06 for CGT group) but still did not reach significance (p=0.64). We acknowledge this trial was not designed or powered to show this effect but felt it important to highlight differences in outcomes between modalities as it has implications for future trials and subsequently how best to administer supplemental therapy.

The potential differential effect favoring robotic therapy is particularly striking given that, though not by design, the Lokomat® group received less overall number of supplemental sessions than the CGT group. The Lokomat® training group had approximately 30% fewer sessions per IRF stay, but the stepping intensity of the gait specific training was higher with this device compared to the CGT intervention. This assessment of increased stepping is subjective, based on observation of therapy sessions; increased step count in robotic therapy compared to CGT was not documented in the current work but has been previously shown [19]. CGT was delivered manually by one or multiple therapist – and so required all participants to coordinate efforts to get productive “training” steps to be realized. Lokomat® training was delivered by the robot and so once the setup was accomplished, the gait training component was more time-condensed and is suspected to have resulted in greater overall steps during the allotted time. Preliminary evidence exists that patients with stroke can improve their walking recovery and quality of life when higher doses of aerobic and stepping activity are provided within 1-4 weeks post injury [28]. Our work similarly, albeit subjectively, supports that increased stepping achieved in the Lokomat® (vs. CGT) may be a key factor in improved outcomes with fewer visits.

Recent work, including reviews and meta-analyses of a moderate body of previous work also suggests but does not clearly indicate differential benefits of robotic training for patients in the acute and subacute phases of stroke [21, 29,30,31,32]. There appears to be more support for robotic therapy assisting non-ambulatory acute stroke patients [21, 32, 33] or generally more impaired patients [34, 35] or favoring acute/sub-acute as opposed to chronic stroke patients [30, 31, 36] although limited evidence exists that higher functioning ambulatory patients can benefit [8]. Our results add evidence to further the idea that robotic based training is helpful earlier post-stroke but even for patients who are ambulatory at baseline. Just as in our study, combination of robotic and conventional therapy compared to the same intensity (usually just meaning duration but not truly intensity) of only conventional found significant improvements in functional ambulation for both training groups, but showed no significant differences between the two intervention groups [37]. That study used mostly different outcome measures than in our study. Gait symmetry and lean body mass improved, however, in the robotic group only. There is growing evidence that combining conventional and robotic modes maybe, in and of itself, a useful approach [23, 31, 32, 38]. Our robotic intervention group actually received this blended therapy as they retained their mandated standard of care that was delivered conventionally while receiving the robotic supplemental training. Overall, optimizing therapy (mode, frequency and dose) needs additional study to help determine for what stage (acute vs. chronic), when specifically during recovery (ex. 30, 60, 90 days post injury, etc.), what functional level (ambulatory vs. not) of stroke patients and how modality can best be used in gait training [10, 39, 40].