mHealth Impact on Gait and Dynamic Balance Outcomes in Neurorehabilitation: Systematic Review and Meta-analysis

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for systematic reviews of randomized controlled trials [32] was followed to report the present systematic review and meta-analysis (Online resource 1). The search was performed up to April 2023 in the following databases: PubMed, Web of Science, Scopus, and Physiotherapy Evidence Database (PEDro). It was adapted for the rest of the databases from an initial search of the PubMed database (Online resource 2). No filters were used in terms of language or publication date.

The inclusion criteria were defined according to the PICOS framework [33] (Population, Intervention, Comparison, Outcomes, Studies): i) RCT including adults (over 18 years old) with neurological disorders, ii) comparing mHealth systems (smartphones or tablets) with a control group, iii) mHealth systems used during gait and balance training, apps based on physical activity (PA) programs in real-world settings, and mobile apps providing feedback to enhance self-patient management and participation to rehabilitation programs, iv) Control group was based on conventional therapy (physical and/or occupational therapies, such as specific gait and balance training, home-based PA programs and active behavior recommendations) and/or not intervention, v) gait and dynamic balance as outcomes.

The exclusion criteria were: i) Systems based on videoconferencing platforms (e.g. Skype, Zoom, etc.), virtual reality glasses, or websites, ii) Studies related to cost-effectiveness and usability of mHealth systems.

Two authors (M.M.L and J.A.M.M.) retrieved and screened the studies by the titles and abstracts independently, removing duplicate articles and identifying potentially relevant articles according to the pre-established criteria. Subsequently, full texts of the retrieved records were downloaded and analyzed for final inclusion in the systematic review. An additional reviewer (D.L.A.) was considered for consensus when needed. Some information was extracted to detail the studies characteristics: authors, year of publication, country, characteristics of the participants (number of participants in both groups, mean age, neurological disorder), and the intervention characteristics (types, session length and frequency, total duration, outcomes measured, measurement instruments, and results). In addition, each mHealth system employed as an intervention was identified, specifying its purpose and main features.

The PEDro scale [34] was used to evaluate the methodological quality of the studies. A study scored as 6 or more is considered having good methodological quality (6-8: good; 9-10: excellent) and scored as 5 or less is considered being of acceptable or poor quality (4-5: acceptable; <4: poor) [35]. The Cochrane Risk of Bias (RoB) Collaboration’s 2.0 tool [36] was used for assessing the risk of bias. This evaluation covers five domains of bias: (1) bias arising from the randomization process, (2) bias due to deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in measurement of the outcome, and (5) bias in selection of the reported result. The RoB of each domain was categorized as: “low risk”, “some concerns”, and “high risk”. Also, an overall RoB assessment of the RCT was performed according to the recommendations in the guidance document. Those assessments were performed by two authors (M.M.L and J.A.M.M.) independently. When there was a difference in the scores between the authors, the final decision was discussed with a third author (D.L.A.).

Several meta-analyses were performed with The Review Manager 5.4 software to summarize the comparisons of the use of mHealth systems (experimental group) versus a control group. The studies were classified based on the measuring instrument to assess gait or dynamic balance. The effect measure was the mean difference (MD), reported along with their 95% confidence intervals (CI). The instructions of the Cochrane Handbook for Systematic Reviews of Interventions for obtaining standard deviation from confidence intervals and change-from-baseline standard deviation were employed when needed. After checking for homogeneity with the chi-square test and the I² statistic (0-40% might not be important; 30-60% may represent moderate heterogeneity; 50-90% may represent substantial heterogeneity; and 75-100% considerable heterogeneity) [37], a fixed-effect model was used when no heterogeneity was detected, and a random model was used otherwise. The significance level was set at alpha=0.05, and the results were presented in forest plots.

The average age was 60.4, although three documents did not provide information [39‐41]. All studies included both genders and the most common disease was stroke, studied in eight articles [38, 42‐47] (Table 1).

Regarding the mHealth systems, they can be classified as those specifically focused on gait training, balance training or those based on home-based PA programs. First, gait training was performed by four different apps: i) ZyMiMetronome App [47] and GotRhythm App [39], two metronome apps that provided a rhythmic auditory stimulation during gait training; ii) Virtual Active App [44], used for speed-interactive treadmill training providing visual feedback through filmed world-famous landscapes; and iii) Audio-biofeedback (ABF)-gait and Freezing of Gait (FOG)-cue Apps [41], part of the CuPiD system, provided biofeedback by verbal instructions and visual information, respectively, during gait recovery. Second, for balance training, a group of apps called CSMi [45] was used, included in the SPVFTCT (smartphone-based visual feedback trunk control training) system, designed for trunk control training. It allowed real-time feedback through the monitoring of weight shifts during the execution of the proposed goal task. Similarly, the Compass Inclinometer App [42] was used for providing patients visual feedback while performing a home-based sitting balance training program. Another mHealth system used for balance training was the Farmalarm App [43], which guided home-based core stability exercise (CSE) through description, photo, and video of each exercise, and recorded its performance. Finally, we found apps providing exercise programs to enhance PA. They were not specifically designed for training gait and balance, but the impact of these apps on these outcomes was assessed. The apps used were Wellpepper App [48], Fitlab Training App [46], Lose it! [49], iPro Habit Tracker Apps [49], Memories App [49], Pt Pal App [40], Patient Information App (PIA) [50], and ad-hoc app [38]. They provided home-based PA programs including videos and prescription information, allowing the monitoring (behaviors, symptoms, or goal achievement), the guidance of the PA performance (visual and/or auditory), and the communication with their physical therapist. More specifications about the mHealth systems employed are shown in Table 2.

Regarding the control interventions, conventional physical and occupational therapies were included, such as specific gait and balance training, home-based PA programs, and active behavior recommendations. None included study compared the mHealth group with no intervention or waitlist.

Concerning the measuring instruments, several motion analysis systems were used (GAITRite platform [47], OptoGait system [44], and ProtoKinetics platform [41]) for gait spatiotemporal analysis. Furthermore, the 10-meter Walk Test (10MWT) [40, 46] was also used for the gait speed analysis. For gait and dynamic balance analysis, the Timed Up and Go (TUG) test [40, 45‐47], the Four Squared Step Test (FSST) [41, 47], the 6-minute Walk Test (6MWT) [40, 46, 48‐50], the 2-minute Walk Test (2MWT) [41, 50], and the mobility subscale of the Postural Assessment Scale for Stroke (PASS) [42, 43] were used.

The overall PEDro score was 6.23 (range 3-8), implying a moderate quality of the articles included in this review. Ten studies [38, 40‐45, 48‐50] obtained a good methodological quality, two were acceptable [46, 47], and one was poor [39] (Online resource 3). Concerning the assessment of the RoB by domain, 7 studies had low RoB and the rest had some concerns for the allocation domain. For the second domain (bias due to deviations from intended interventions), 9 studies had low RoB while 4 studies had some concerns. In case of third (missing outcome data) and fourth (measurement of the outcomes) domains, 11 and 10 studies had low RoB, respectively. For the last domain, in the selection of the reported results, 11 studies had low RoB and only one study had some concerns (Online resource 4). Regarding the overall judgement, 5 studies had a low RoB, 5 studies had some concerns, and 3 studies had a high RoB, as shown in Online resource 4.

The mean, standard deviation and sample sizes of the different subgroups and measures, along with the MD (95%CI), are shown in Figs. 2 and 3. No statistical heterogeneity was observed in seven of the ten outcomes studied. However, stride length and FSST presented moderate heterogeneity (I²=49% and I²=47%, respectively), while PASS (mobility) presented substantial heterogeneity (I²=75%). We found significant differences in temporal gait parameters, speed (MD=0.10;95%CI=0.07,0.13; p<0.001) (Fig. 2A) and cadence (MD=8.01;95%CI=3.30,12.72; p<0.001) (Fig. 2B). Regarding spatial gait parameters, the stride length results were no significant (MD=0.08;95%CI=-0.00,0.16; p=0.06) (Fig. 2C). Conversely, there were significant differences in affected step length outcome (MD=8.89;95%CI=4.88,12.90; p<0.001) (Fig. 2D) and non-affected step length (MD=8.08;95%CI=2.64,13.51; p=0.004) (Fig. 2E). Concerning the dynamic balance, significant differences were observed in the TUG test (MD=-7.15;95%CI=-9.30,-4.99; p<0.001) (Fig. 3A), and PASS (mobility) (MD=1.71;95%CI=1.38,2.04; p<0.001) (Fig. 3B). The FSST (MD=-1.90;95%CI=-3.82,0.03; p>0.05) (Fig. 3C), 6MWT (MD=11.74;95%CI=-16.40,39.89; p=0.41) (Fig. 3D), and 2MWT (MD=5.13;95%CI=-10.55,20.81; p=0.52) (Fig. 3E) results were no significant. Therefore, the results of the present study suggest that the use of mHealth systems was slightly more effective than the conventional rehabilitation of training and enhancing gait and balance outcomes according to motion analysis systems, the TUG test, and the PASS (mobility).

Although this review and meta-analysis provide an evidence-based overview of the use of mHealth in subjects with neurological disorders, some limitations should be remarked. Our results should be taken with caution because of the heterogeneity of neurological disorders and the variety of assessment tools used. Likewise, several intervention protocols were used in experimental and control groups. These facts could influence the comprehension and interpretation of findings. Thus, protocols standardization is needed for future studies. Another common limitation, along with the studies included, was the lack of assessor blinding for group allocation, resulting in a high risk of bias in this domain. As well as stated previously, the sample size involved in the studies was limited, making the generalization of the obtained results difficult. Finally, it is necessary to perform trials with long-term follow-up to assess whether the mHealth effects are maintained in neurorehabilitation.

Nevertheless, despite of those limitations, this systematic review and meta-analysis provides the first evidence-based findings about the use of mHealth systems for gait and balance recovery in neurorehabilitation. Therefore, this manuscript could be used as the base of future clinical studies and discover the necessity of further research that will allow the stratified analysis of the results.