Stroke is a major global health concern, ranking as the second leading cause of death and the third leading cause of death and disability combined (as expressed by disability-adjusted life-years [DALYs] lost) in the world. The burden of stroke has increased by 70% in the last three decades, with a global incidence of 13.8 million and an annual death toll of 5.5 million. ^{1, 2} In 2019, stroke accounted for 5.7% of the global disease burden. ¹ In Finland, stroke is the fourth leading cause of DALYs, with an incidence of 182 per 100,000 people and a prevalence of 1686 per 100,000 people in 2019. ³ The economic burden of stroke is substantial, amounting to €745 billion globally (1.12% of the global gross domestic product, GDP). In Finland, the healthcare cost of stroke in 2017 was €640 million (3.1% of total healthcare costs), with a total DALYs cost of €1.1 milliard (0.5% of the GDP). ⁴ These costs include immediate medical care, hospitalization, medications, and rehabilitation services, emphasizing the importance of post-stroke rehabilitation in achieving better functional outcomes and reducing long-term disability.

Motor impairment of the lower extremities is one of the most important determinants of long-term disability. ⁵ Conventional rehabilitation of stroke includes physical therapy among speech and occupational therapies. Part of physical therapy is gait rehabilitation with the guidance of physiotherapists and sometimes the use of a treadmill. Compared to conventional rehabilitation recent literature highlights the positive impact of robot-assisted gait training (RAGT) on functional outcomes in patients with severe stroke. ⁶ RAGT has demonstrated effectiveness in improving gait parameters, including walking speed, step length, and symmetry. ⁷ Furthermore, this technology provides the therapist with objective measures of the patient’s performance during the therapy and enables targeted intensive, task-specific training in a controlled environment to individual patient needs. RAGT is a more effective mode of rehabilitation for certain patient groups, e.g., individuals with severe to moderate impairments in walking abilities and functional mobility than conventional approaches. ^{4, 6} However, other studies, including the LEAPS trial by Duncan et al. and meta-analyses by Merholz et al. and Yamamoto et al., have questioned the superiority of RAGT compared to conventional training. ^{8– 10} These mixed findings highlight the need to understand better which patient groups may benefit most from RAGT. Therefore, more research is needed to define the effectivity of RAGT. ¹¹

Although RAGT therapy is used increasingly, there is still a knowledge gap between RAGT's efficacy, i.e., performance under ideal and controlled circumstances and its effectiveness, i.e., how well it works in normal clinical settings. These data are important as the economic costs for post-stroke care are substantial. ⁴

Real-world data (RWD) generates additional evidence to that found in clinical research settings. Robust RWD can bridge the gap between research and clinical practice to better understand the clinical profile of patients as well as the value of using different treatment options. With this study, the aim was to increase understanding of RAGT for stroke patients based on retrospectively collected RWD from Soite, one of the 21 wellbeing services counties of Finland. The secondary aim was to describe how RAGT was administered in a real-world hospital-based setting as well as to extrapolate the potential benefits of rehabilitation robots in broader healthcare contexts in Soite.

In the past two decades, robot-assisted rehabilitation after stroke has been studied as a promising approach associated with improved functional outcomes and quality of life for individuals recovering from stroke. Our study undertook a thorough evaluation of the effectiveness of RAGT by integrating multiple key findings.

Firstly, step count and walking distance increased over the course of rehabilitation, while body weight support decreased, suggesting improvements in mobility and ambulation among patients receiving RAGT.

This tangible progress is particularly significant in the context of stroke recovery, where the restoration of ambulatory function is a pivotal goal. This finding is consistent with previous research that has shown that RAGT rehabilitation can improve substantially walking ability in people with stroke. ^{14, 15} Stroke survivors prefer improvements in their walking distance rather than their walking speed. ¹⁶ In consistent to our data, a recent study has shown that intensive walking therapy can increase the step count during physical therapy substantially. In Klassen et al. study, the mean step count in the conventional rehabilitation was 580 steps per session, whereas in the intensive therapy group mean step count was over three-fold higher. Moreover, walking endurance benefits achieved were retained over the first year poststroke. ¹⁷ Especially, RAGT in combination with conventional rehabilitation has shown better results than conventional rehabilitation on its own. ^{14, 18}

FIM scores improved clinically significantly during the rehabilitation period in patients receiving RAGT. According to the literature the minimal clinically important difference (MCID) for FIM score is 22 (total FIM), 17 (motor FIM), and 3 (cognitive FIM). ¹⁹ The improvements observed in this study, 39 points for the total FIM score and 35 points for the motor FIM score, were two-fold compared to the MCID. The cognitive FIM change was a little less than the proposed MCID. This is most likely due to higher cognitive scores at the baseline measurement which can be a reason for changes that are below MCID. ¹⁹ Sufficient motor and cognitive functions are important measures of functional independence, and our findings suggest that RAGT may be associated with substantially improved ability to perform activities of daily living. Other studies have also found that RAGT is effective in improving e.g., FIM and other scores that measure gait independence. ^{20, 21} Consistent with our data, Peters et al. found that the high-intensity gait training may improve participants’ cognitive function modestly more effectively than the usual care. ²²

We found that guidance forces were applied asymmetrically between limbs, often providing more support to the hemiplegic side. This was an intentional clinical strategy to encourage symmetry and assist the impaired side while minimizing compensatory reliance on the unaffected side. This is supported by the data indicating that stroke is more common in the left hemisphere, and thus, more guidance force is needed for the right lower limb. ^{23– 25} This is likely because the left hemisphere is responsible for motor control, and damage to the left hemisphere can lead to more severe walking impairments. van Vliet et al have highlighted the importance of targeting rehabilitation interventions more specifically to individual stroke patients. ²⁶ RAGT is a preferred model as it allows this kind of rehabilitation tailored to the individual needs.

Notably, a large proportion of patients with severe stroke receiving RAGT were discharged from hospital to home rather than to assisted living facilities. This outcome may reflect a potential cost-effectiveness of RAGT and the association between RAGT and increased patient independence and self-sufficiency, although causal inference cannot be drawn due to the study design. One study has found that of all stroke patients 78% were discharged to home and 22% to assisted living facilities. ²⁷ That study had included also patients with mild strokes and therefore the likelihood of being discharged to assisted living facilities was even lower compared to our study. ²⁷

In future more research, especially from randomized clinical trials, is needed to define the effectivity of RAGT. Robust RWD can bridge the gap between research and clinical practice to better understand the clinical profile of patients as well as the value of using different treatment options. Future research should investigate the long-term effects of RAGT on walking ability and function in people with stroke. Additionally, it is recommended that future real-world evidence research should include a bigger patient population to investigate the effectiveness of RAGT compared to other rehabilitation methods for stroke patients. A larger and more diverse patient population should be considered to thoroughly evaluate the comparative effectiveness of RAGT in contrast to alternative rehabilitation methods for individuals recovering from stroke. Additionally, future research employing a randomized controlled trial with a control group is necessary to confirm the observed trends and establish definitive causal relationships.

It is essential to acknowledge certain limitations inherent in our study. Firstly, the number of patients with stroke receiving RAGT was notably small, which may impact the generalizability of our findings to a broader population. Additionally, due to retrospective data limitations, stroke subtype (ischemic vs haemorrhagic), NIHSS scores, cognitive assessments, and comorbidities were not consistently recorded or able to be retrieved from the EMR systems. Future studies should aim to include this information to better characterize treatment response and subgroup differences.

The average session duration in this study was shorter than the 45–60 minutes commonly recommended in the literature. This reflects real-world clinical constraints, such as patient fatigue, scheduling limitations, and availability of trained staff. Furthermore, patients in the subacute phase may tolerate only shorter durations in body weight support level exceeded 50%, which is above the typically recommended threshold. This was necessary due to the severity of motor impairment and early stage of recovery in many patients. Although higher support levels may reduce active muscle effort, they were essential to ensure patient safety and facilitate consistent stepping patterns early in therapy.

Furthermore, our study exclusively draws data from one wellbeing services county in Finland, potentially limiting the broader applicability of our results to a more diverse range of settings or regions. We also observed substantial variability in the number of RAGT sessions per patient (ranging from 2 to 15). This likely reflects clinical judgment, individual tolerance, and recovery trajectory. Future research could examine whether greater session frequency correlates with improved outcomes in a dose-response model.

To thoroughly assess the effectiveness of RAGT, it would be valuable to compare outcomes with a control group or baseline measures to isolate the impact of the robotic intervention beyond typical recovery. Additionally, analysing patient characteristics and concurrent interventions can offer deeper insights into how RAGT influences post-stroke rehabilitation outcomes. These factors should be considered when interpreting and applying the implications of our study findings. Additionally, session parameters such as speed, body weight support, and guidance force were determined dynamically by the therapist rather than a fixed algorithm. While this approach enhances patient individualization, it may limit reproducibility and generalizability of the findings.

In conclusion, our study has several important implications for the treatment of stroke patients. It suggests that early RAGT after stroke can improve walking ability, function, and discharge to home. These findings collectively advocate for the incorporation of RAGT into comprehensive stroke rehabilitation programs, paving the way for improved patient outcomes and post-stroke recovery trajectories.

According to Finnish legislation, registry-base studies require a research permit from the health authority who owns the registry (Act on the Secondary Use of Health and Social Data; 552/2019, Finland). Ethical approval from ethics committee is needed only for interventional studies (Medical Research Act; 488/1999, Finland). Therefore, ethical approval was not required for the study. The study was granted a research permit approved by the Wellbeing Services County of Central Ostrobothnia on 9.1.2023. They also waived the need for ethical approval and informed consent, given the retrospective register-based study design. We confirm that all methods were carried out in accordance with relevant guidelines and regulations.