Journal of Rehabilitation Medicine 51-2 | Page 9

84 J. Schröder et al. tency of results into account, there is level B evidence for improved walking independence after repetitive gait training. • Walking speed: This was assessed in 9 RCTs and pooling yielded a non-significant homogenous SES (9 RCTs; n = 672 [exp 342; ctr 330]; MD = 0.05 [fixed]; 95% CI –0.00 to 0.11; p = 0.06; I 2  = 42%). No significant subgroup difference was found (p = 0.41). • Walking endurance: In 4 RCTs, endurance was mea- sured. When pooling data, a significant homogenous SES (4 RCTs; n = 406 [exp 206; ctr 200]; MD = 24.36 [fixed]; 95% CI 3.58–45.14; p = 0.02; I 2  = 43%) is found. Taking the inconsistency of results into ac- count, there is level B evidence for improved walking endurance after repetitive gait training. If pooling end-effector trials only, a homogeneous significant SES is calculated (3 RCTs; n = 309 [exp 154; ctr 155]; MD = 32.08 [fixed]; 95% CI 8.30–55.86; p = 0.008; I 2  = 44%). Body function level: • Motor control: Four RCTs assessed the FM-L and pooling results yielded a non-significant homo- geneous SES (4 RCTs; n = 179 [exp 91; ctr 88]; MD = 0.52 [fixed]; 95% CI –1.52 to 2.59; p = 0.62; I 2  = 28%). If pooling exoskeleton trials only, a non- significant heterogeneous SES is calculated (3 RCTs; n = 119 [exp 63; ctr 56]; MD = 0.76 [fixed]; 95% CI –1.83 to 3.36; p = 0.56; I 2  = 51%). • Muscle strength: For a comparison on the MI-L, results of 5 RCTs could be pooled. This resulted in a non-significant heterogeneous SES (5 RCTs; n =  364 [exp 185; ctr 179]; MD = 3.64 [random]; 95% CI –2.88 to 10.17; p = 0.27; I 2  = 56%). If isolating effects of end-effector trials a significant homogeneous SES (3 RCTs; n = 230 [exp 113; ctr 117]; MD = 8.00 [random]; 95% CI 2.08–13.93; p = 0.008; I 2  = 8%) is identified. DISCUSSION No between-group difference in the occurrence of adverse events and drop-outs was found despite the demanding nature of the intervention. This suggests it is feasible to provide repetitive training early after stroke by the use of an overhead harness system for support of body weight and manual or electromecha- nical assistance in forward progression of the paretic leg. Statistical significant effects on walking indepen- dence (at follow-up) and endurance is found in favour of repetitive training, according to level B evidence. Sub-analyses revealed that these effects are based mainly on studies investigating RAGT provided with an end-effector robot. www.medicaljournals.se/jrm Dose-response relation in stroke rehabilitation In the context of neurological rehabilitation, repetitive training leads to task-specific improvements (10, 31) and associated neuroplastic re-organization (50) if a sufficient dose of practice is provided. In animal mo- dels, synaptic changes in the motor cortex are observed after 400, but not after 60 reach-movements (51) and gait training is effective only if at least 1,000 steps are performed during a treadmill session (52). Correspon- ding findings in clinical research are in favour of a dose-response relation in stroke rehabilitation (17, 53). Despite this solid association between larger quantities of practice and greater gains, inpatient rehabilitation is described as a time of being physically inactive (54, 55) and the practice dose is far less than is provided in previously mentioned stroke models: patients walked for a mean of 250 (21) steps, while non-ambulatory patients walked for as little as 6–16 steps (56) during a therapeutic session aiming at gait recovery. Technological advances can be of great value in providing more intensive rehabilitation, as robots let non-ambulatory patients train at much higher doses (57). For example, the Gait Trainer allows patients to execute approximately 1,000 steps in a session, while assistance of a single therapist is usually sufficient (35, 39, 43). In line with a dose-response relation, training with such a device appears effective in improving walking independence and endurance. This implies the importance of practice repetitions when designing effective interventions (58) and, in more general terms, the significance of motor learning in stroke rehabilita- tion (14, 50, 59). However, the dose-response relationship is not linear, indicating that other factors have an influence (53). Morone et al. (43, 44) compared responsiveness to training between groups who differ in baseline sco- res on the MI-L (MI-L≈16 vs 52). Outcomes clearly demonstrate that only the more impaired patients be- nefit (43, 44), which is supported by Pohl et al. (35) as they found impressive effects in a more affected population (MI-L≈32; see Table II). Interestingly, the initial muscle strength of the paretic leg (e.g. assessed with the MI-L (6)) measured within the first days to weeks post-stroke is associated with walking ability at 6 months (5, 6, 60). Therefore, it appears that robot- assisted training was most effective in those patients with a poor prognosis. This might be related to a greater treatment contrast, since the more affected patients do not engage in intensive rehabilitation under conven- tional conditions (56). As suggested by Morone et al. (57) future research should not investigate if RAGT is effective, but rather who may benefit (43, 57).