Effects of robotic upper limb treatment after stroke on cognitive patterns: A systematic review

2.1 Data sources and searches

PubMed and Web of Science were used to perform the literature search. The electronic search was conducted on March 2021. To avoid missing some key studies, cross-referencing was used from each publication obtained through the electronic search. The search strategies, combining relevant search terms with Boolean operators (OR/AND), are listed in Table 1, for the two databases, separately. The protocol was registered on PROSPERO (no CRD42021288946). The Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement was followed in the reporting.

In agreement with the PRISMA guidelines (Page et al., 2021), the inclusion criteria were set following the PICOTS-SD (Brown, 2020) (participants, intervention, comparisons, outcomes, timing of outcome measurement, settings, study design) strategy, as reported below:

•

Participants (P)=persons with stroke;

•

Intervention (I)=upper limb robotic rehabilitation focused on the recovery of upper limb motor function;

•

Comparisons (C)=conventional treatment, no treatment, or treatment with other robotic or technological devices or healthy participants;

•

Outcome (O)=upper limb motor function measured by clinical scales or instrumental parameters;

•

Timing of outcome measurement (T)=before and after the intervention;

•

Settings (S)=inpatient or outpatient settings;

•

Study design (SD)=randomized controlled trial (RCT).

We therefore included RCT an patients with stroke comparing an upper limb robotic rehabilitation with other interventions (such as conventional treatment, or other devices) focused on the recovery of upper limb motor function measured by clinical scales or instrumental parameters. RCTs with two or more arms were considered. If studies did not use robotic devices or were not focused on stroke patients, they were excluded from this review.

2.3 Study selection and data extraction

Two independent reviewers evaluated the studies retrieved from the electronic search based on the titles and abstracts of the studies. After this preliminary screening process, the full text of all eligible studies was analyzed and independently evaluated to determine whether or not they met the inclusion criteria. A third reviewer was brought in to resolve any disagreements on the study’s eligibility between the two reviewers. The flow diagram of the article selection procedure is reported in Fig. 1. After inclusion, the study characteristics, research goals, and main findings were extracted and summarized. Specifically, the extracted information included: total number of patients randomized, mean time since the acute event of the enrolled patients (classified as lower or higher than 6 months), description and dose of the intervention in the experimental group, description, and dose of the intervention in the control group(s), primary and secondary outcomes. The PEDro scale (Cashin and McAuley, 2020) was used to assess the methodological quality of the studies.

OPEN IN VIEWER

Fig. 1

PRISMA 2020 flow diagram for new systematic reviews which included searches of databases, registers and other sources.

Moreover, in line with the goal of this review, the following data were considered and analyzed: presence of cognitive inclusion criteria (with description, when applicable), cognitive outcome measures (with description, when applicable), and any investigation of the relationship between motor and cognitive outcomes (with description, when applicable).

3.1 Data synthesis

We found 880 records through the research method. After duplicates were removed, articles were screened for title and abstract, and 90 records were assessed for eligibility.

After full-text reading, 81 studies were included in the qualitative analysis of this systematic review. Figure 1 reports the flowchart of the research.

Supplementary Table 1 summarizes the characteristics of the included studies. According to the inclusion criteria, all reports are RCTs published between 2000 (Volpe et al., 2000) and 2021 (Ambrosini et al., 2021; Chinembiri et al., 2021; Doost et al., 2021; Gueye et al., 2021; H. C. Lee et al., 2021; Park, 2021).

3.2 Population

The studies included a total population of 3922 stroke patients. The sample size varied from 12 (Brokaw et al., 2014; Daly et al., 2005; Iwamoto et al., 2019) to 770 (Rodgers et al., 2019a).

Time onset varied in according to studies’ inclusion criteria: twenty-nine studies included only patients whose time since the stroke event was under 6 months; thirty-three studies included only chronic stroke patients while the remaining studies included both phases. Time between stroke onset and start of treatment was specified in 67 studies and it ranged between 14 days (Gueye et al., 2021; Volpe et al., 2000) and 9 months (Park, 2021). Supplementary Table 1 reports population characteristics.

3.3 Intervention

Robot upper limb trainingwas used in the intervention group of all included studies. Complete treatment characteristics are reported in Supplementary Table 1.

3.4 Robotic treatment characteristics

All studies have at least one group that underwent RT. According to the treated district, in 18 studies (Abdullah et al., 2011; Chinembiri et al., 2021; Cho & Song, 2019; Fazekas et al., 2007; Han Yoo et al., 2012; Horsley et al., 2019; Iwamoto et al., 2019; Kahn et al., 2006; Klamroth-Marganska et al., 2014; M. Lee et al., 2018; Park et al., 2020; Perini et al., 2021; Qian et al., 2017; Rodgers et al., 2019b; Serrezuela et al., 2020; Takahashi et al., 2016; Timmermans et al., 2014; Tramontano et al., 2020) the entire upper limb was treated; shoulder was rehabilitated in five studies (Burgar et al., 2011; Cho and Song, 2021; Kim et al., 2019; Lum et al., 2002; Straudi et al., 2020); only Mazzoleni et al. (2019) considered wrist rehabilitation, while hand rehabilitation was considered in 13 studies (Ang et al., 2014; Calabrò et al., 2019; Grigoras et al., 2016; Y. Huang et al., 2020; Hwang et al., 2012; H. C. Lee et al., 2021; Orihuela-Espina et al., 2016; Park, 2021; Ranzani et al., 2020; Sale et al., 2014; Susanto et al., 2015; Vanoglio et al., 2017; Villafañe et al., 2018). None of the studies rehabilitated only elbow or forearm joints.

The remaining studies combined the treatment of two or more different districts: 20 (Ambrosini et al., 2021; Carpinella et al., 2020; Daly et al., 2005, 2019; Dehem et al., 2019; Doost et al., 2021; Gandolfi et al., 2019; K. W. Lee et al., 2016; Lee KW et al., 2017; S. H. Lee et al., 2020; Lo et al., 2010; Lum et al., 2002; Masiero et al., 2007, 2014; McCabe et al., 2015; Rabadi et al., 2008; Rosenthal et al., 2019; Tomić et al., 2017; Volpe et al., 2000; Xu et al., 2020) rehabilitated shoulder and elbow; robotic rehabilitation for forearm and wrist was considered in eleven studies (Hesse et al., 2005, 2014; Housman et al., 2009; Hsieh et al., 2011, 2016; Hsu et al., 2019; C. S. Hung, Hsieh, Wu, Chen, et al., 2019; C. S. Hung, Hsieh, Wu, Lin, et al., 2019; C. Shan Hung et al., 2019; Liao et al., 2012; Wu et al., 2012); wrist and hand was treated in two studies (Kutner et al., 2010; Wolf et al., 2015); Qian et al. (2019) used a robotic device for elbow, wrist and hand rehabilitation, while five studies (Daunoraviciene et al., 2018; Gueye et al., 2021; Jiang et al., 2021; Taveggia G et al., 2016; Zengin-Metli et al., 2018) underwent RT for shoulder, elbow and hand.

Conroy et al. (2019) combined two robots, one for shoulder and elbow and the other for wrist, as well as Edwards et al. ((2019); Aprile et al. (2020) used different robotic devices for the treatment of the upper limb, treating shoulder, elbow, wrist and hand in different phases; Conroy et al. (2011) rehabilitated, with robotic devices, shoulder and elbow in one group and forearm and wrist in another, while Brokaw et al. (2014) combined a robot for shoulder rehabilitation with one for hand rehabilitation.

Robotic characteristics (i.e., name, industry and country of production were specified in all studies except seven (Abdullah et al., 2011; Grigoras et al., 2016; Y. Huang et al., 2020; Qian et al., 2017, 2019; Serrezuela et al., 2020; Susanto et al., 2015) that used their own robotic prototypes to treat patients enrolled. Bi-Manu Track (Reha-Stim Co., Berlin, Germany) (Hesse et al., 2003) was the most used robotic device, followed by Armeo Spring (Hocoma AG, Volketswil, Switzerland) (Armeo®Spring –Hocoma, n.d.), Amadeo (Tyromotion, Graz) (Balasubramanian et al., 2010), Mit Manus (Interactive Motion Technologies, Cambridge, MA) (Aisen et al., 1997) –also described as InMotion2 (Interactive Motion Technologies, Inc., Cambridge, MA, USA) (Krebs et al., 1998).

Except for Brokaw et al. (2014) and Kutner et al. (2010) – who reported the total time of RT – and Rosenthal et al. (2019) –who didn’t specify session duration – all studies specified RT duration characteristics (i.e., total of sessions, sessions per week, duration of each session). RT ranged from 2 to 64 sessions, with an average of 4 sessions per week. Duration of each RT varied from 20 to 120 min.

Therefore, patients performed conventional therapy, propaedeutic therapy, or additional therapy (i.e., functional electrical stimulation –FES) in more than half of the studies. Characteristics of RT and additional therapy are reported in Supplementary Table 1.

All studies have at least one control group. All the studies control groups are composed of patients who have the same baseline characteristics as the robotic group, except for Doost et al. (2021), who used a group of healthy subjects to normalize the obtained data.

In sixty studies patients in the control group underwent conventional therapy, of these in three studies (Burgar et al., 2011; Masiero et al., 2007; Volpe et al., 2000) patients were exposure to the robot in addition to the traditional treatment.

Regarding the remaining studies, in fifteen studies, comparison underwent two different types of RT: Iwamoto et al. (2019) and Conroy et al. (2019) combined RT and conventional therapy.

Therefore, in four studies (Daly et al., 2005, 2019; Hesse et al., 2005; McCabe et al., 2015) functional neuromuscular electrical stimulation (FNS) or FES was used in control group. Characteristics of comparison population and their rehabilitation programme are reported in Supplementary Table 1.

3.6 Cognitive inclusion/exclusion criteria

A total of seventy-three article included cognitive inclusion or exclusion criteria.

Nineteen studies looked at both cognitive inclusion and exclusion criteria, twenty-nine looked at only inclusion criteria, and twenty-five looked at cognitive exclusion criteria.

A total of forty-nine studies considered cognitive inclusion criteria and thirty-seven used a single or more scales to evaluate cognitive inclusion performances.

The Mini Mental State Examination (MMSE) (Folstein et al., 1975) was the most used cognitive screening (twenty-nine studies out of forty-six studies), however the cut off varied based on the studies: 24 points was the cutoff longer used (Calabrò et al., 2019; Gandolfi et al., 2019; Han Yoo et al., 2012; Hsieh et al., 2011; Hsu et al., 2019; Hung CS et al., 2019; Kutner et al., 2010; Perini et al., 2021; Tramontano et al., 2020), followed by 22 (Hsieh et al., 2016; C. S. Hung, Hsieh, Wu, Chen, et al., 2019; C. Shan Hung et al., 2019; Liao et al., 2012; Wu et al., 2012), 21 (Daunoraviciene et al., 2018; Y. Huang et al., 2020; Qian et al., 2017, 2019; Susanto et al., 2015) (50,68,73,104,106), 20 (Ambrosini et al., 2021; Sale et al., 2014), 18 (Jiang et al., 2021; Masiero et al., 2007). Only Dehem et al. (2019) used a cut-off of 15 points, but it was also required that patients had the ability to “understand instructions”. Therefore, Rosenthal et al. (2019) did not specify the cutoff utilized.

Five studies specified the use of Korean version of MMSE (Park and Kwon, 1989) with different cut offs according to each study: three studies (Cho and Song, 2019, 2021; Park, 2021) settled it at 24 points; Lee et al. (2018) considered 21 points while Lee et al. (2017) considered patients with a MMSE ≥ 10 points.

Other scales used were: Lowenstein Occupational Therapy Cognitive Assessment (LOTCA) (Hsu et al., 2019), Line Bisection Test and the Korean version of the Motor-free Visual Perception Test—Third Edition (MVPT-3) (Park, 2021), Catherine Bergego Scale (Gueye et al., 2021), Cognistat instrument (Lum et al., 2002) and Short Portable Mental Status Questionnaire (Wolf et al., 2015).

Regarding the exclusion criteria, forty-four studies considered at least one cognitive evaluation. Similar to cognitive inclusion criteria, MMSE (Folstein et al., 1975) is the most used assessment scale, although the cutoff is not standardized: it varied from 20 (Carpinella et al., 2020) to 27 (Orihuela-Espina et al., 2016), however the most commonly used value is 21 (Daunoraviciene et al., 2018; Lum et al., 2006; Straudi et al., 2020; Taveggia G et al., 2016). Therefore, Kim et al. (2019) evaluated their patients using Korean version of the MMSE (Park and Kwon, 1989), with a 15 points cut-off. Another evaluation scale used to evaluate severe cognitive deficit was the Levels of Cognitive Functioning-Revised (Ranzani et al., 2020).

Other assessment scales used could be classified according to the neuropsychologic impairments: the Alexander Scale (Hwang et al., 2012) was the only assessment scale for apraxia; the Bell Test (Masiero et al., 2014) was used to assess attention level, while the Barrage Test (Masiero et al., 2014) and the Star Cancellation Test (Wolf et al., 2015) were used to assess neglect. Regarding aphasia, four assessments were used: the NIH Stroke Scale (question IX) (Hwang et al., 2012), the Neuropsychological Aphasia Test (Masiero et al., 2014), the Gellanza-Coen Test (Masiero et al., 2014) and the Goodglass and Kaplan Test (Ranzani et al., 2020). Supplementary Table 1 reports data on cognitive inclusion and exclusion criteria.

3.7 Cognitive outcome

Seven studies (Daunoraviciene et al., 2018; Gueye et al., 2021; Iwamoto et al., 2019; Park, 2021; Ranzani et al., 2020; Volpe et al., 2000; Zengin-Metli et al., 2018) analyzed cognitive outcomes.

The most used was the Functional Indipendence Measure (FIM) cognitive subscore that was analyzed in three studies (Iwamoto et al., 2019; Volpe et al., 2000; Zengin-Metli et al., 2018); the MMSE was used in two studies (Ranzani et al., 2020; Zengin-Metli et al., 2018). Other measures of cognitive outcomes are: the Addenbrooke’s Cognitive Examination-Revised (ACE-R) (Daunoraviciene et al., 2018), the Frontal Assessment Battery (FAB) (Ranzani et al., 2020) and the Montreal Cognitive Assessment (MoCA) (Gueye et al., 2021).

Unilateral spatial neglet was investigated in two different studies with three different scales: the Albert Test (AT) (Park, 2021; Ranzani et al., 2020), the Line Bisection Test (LBT) (Park, 2021) and the Catherine Bergego Scale (CBS) (Park, 2021).

In all the studies, a clinical cognitive evaluation was performed before and after the treatment. One study (Iwamoto et al., 2019) carried out mid-term pre-treatment and post-treatment evaluations every 5 days, while Ranzani et al. (2020) assessed follow-up evaluations at 8 weeks and 6 months after the end of treatment. Results are shown in Supplementary Table 1.

Significant improvement in between group analysis is shown in ACE-R (Daunoraviciene et al., 2018). FIM cognitive subscore showed significant improvement in both experimental and control groups in two out of three studies (Volpe et al., 2000; Zengin-Metli et al., 2018).

Ranzani et al. (2020)s observed minor improvements –not statistical significant –in both groups over time in FAB score.

Regarding MMSE, Zengin-Metli et al. (2018) found a significant improvement in the control group, maybe because a significant difference was observed between the robot and the control group in terms of pre-treatment MMSE levels (pretraining: 17 normal robotic group vs 8 normal and 6 mild in control group), while no significant improvement was found in Ranzani et al. (2020). Therefore, no significant improvement was found in MoCA results (Gueye et al., 2021).

Regarding unilateral spatial neglet, Park et al. (2021) found that RT reduces hemispatial neglect symptoms in the participants’ activities of daily living: AT and LBT improved significally in experimental group, while CBS showed significant improvement in both groups, but a statistically significant difference in changes in RT group. Instead, Ranzani et al. (2020) observed only minor improvements in both groups over time.

3.8 Methodological quality

Methodological quality was assessed with PEDro scale (Cashin and McAuley, 2020): according to the compilation guidelines of the assessment scale, “eligibility criteria” was not used to calculate the PEDro score. According to the literature, in order to simplify the interpretation of results, articles were classified in four categories: a total PEDro score of 0-3 are considered ‘poor’, 4-5 ‘fair’, 6-8 ‘good’, and 9-10 ‘excellent’. Therefore, for trials evaluating complex interventions (e.g., exercise) a total PEDro score of 8/10 is optimal (Summary of Measurement Properties of the PEDro Scale –PEDro, n.d.).

Details of the PEDro score are reported in Table 2: the lower evaluation was 2/10 (Fazekas et al., 2007) and the higher was 8/10 (Calabrò et al., 2019; Carpinella et al., 2020; Cho and Song, 2019, 2021; Gandolfi et al., 2019; Horsley et al., 2019; Hsieh et al., 2016; Hsu et al., 2019; Y. Huang et al., 2020; C. S. Hung, Hsieh, Wu, Lin, et al., 2019; C. Shan Hung et al., 2019; Klamroth-Marganska et al., 2014; Park, 2021; Perini et al., 2021; Rodgers et al., 2019b; Timmermans et al., 2014; Tramontano et al., 2020; Wu et al., 2012).

Most of the studies (sixty-three over eighty-one) were considered “good”, while there were not “excellent” evaluation.

“Random allocation” was undertaken by all studies while “concealed allocation” was considered only in 28 studies and “baseline comparability” in 71 studies. Most of the studies did not consider the blind of patient and therapists, of all the articles included, six articles (Doost et al., 2021; Edwards et al., 2019; Mazzoleni et al., 2019; Park, 2021; Perini et al., 2021; Wu et al., 2012) and two (Edwards et al., 2019; Wu et al., 2012) of the therapists. Assessors were blind in 66 over 81 studies.

Regarding the statistical analysis, 66 studies had “adequate follow-up”, thirty-nine studies reported “intention-to-treat analysis”, while “between-group comparisons” were reported in 78 studies and “point estimates and variability” were 76 studies.