This study is based on the ICF framework [32], analyzing the intervention effects of wearable devices on gait, balance, and quality of life in patients with PD. The Population, Intervention, Comparison, and Outcome (PICO) structure is detailed in Table 1. This study has been registered in PROSPERO (International Prospective Register of Systematic Reviews) [33], with registration number CRD42024585686. The review protocol is available in the PROSPERO record.

We conducted a comprehensive literature search in PubMed (National Library of Medicine), Embase (Elsevier, Embase.com), the Cochrane Library (Wiley), and Web of Science Core Collection (Clarivate) from inception to November 18, 2025. We also searched ClinicalTrials.gov on November 18, 2025, to identify ongoing or unpublished RCTs evaluating wearable-device interventions in PD.

Databases were searched separately via their native interfaces (ie, no multidatabase searching on a single platform). No search updates were performed after the final search date, and no email alerts were set. No additional online or print sources (eg, journal tables of contents, conference proceedings, or websites) were purposefully searched or browsed.

We screened the reference lists of relevant systematic reviews and all included trials (backward citation searching) to identify additional eligible studies. Reference list screening was performed during full-text assessment and was completed on November 18, 2025. We did not perform forward citation tracking (citing-reference searches) or set up citation alerts. When necessary, we contacted corresponding authors of included studies to request clarification or missing outcome data. No additional information sources or search methods (eg, gray literature databases, preprint servers, or Google Scholar) were used beyond those described above.

The search strategy combined controlled vocabulary terms (eg, MeSH [Medical Subject Headings] and Emtree) and free-text keywords related to PD, wearable or sensor-based devices, and RCTs. We did not use published search filters (eg, validated RCT filters); trial design was captured using a combination of controlled vocabulary and free-text terms. Search strategies were developed de novo and were not adapted or reused from previous literature reviews.

Searches were limited to English-language publications involving human adults; no restrictions were placed on publication status. These restrictions were applied to align with the eligibility criteria (adult PD populations and RCTs evaluating intervention efficacy) and for feasibility of screening and data extraction.

The search strategy and its reporting followed PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses literature search extension). The full search strings for each database and information source (including all search terms, Boolean operators, limits, and the date last searched) are provided in S1 in Multimedia Appendix 1. The search strategy was not formally peer-reviewed.

Two reviewers (JW and WZ) independently screened all retrieved records. In the first stage, titles and abstracts were assessed for relevance and study design. In the second stage, full texts were evaluated against the predefined inclusion and exclusion criteria. Any discrepancies were resolved by discussion; if consensus could not be reached, a third reviewer adjudicated. The study selection process was documented using a PRISMA 2020 flow diagram. The completed PRISMA checklist is provided in Multimedia Appendix 2. No automation tools were used for study selection. As only English-language reports were eligible, no translation was required for screening.

Data were extracted independently by 2 reviewers (MF and NX) using a piloted, standardized extraction form. Discrepancies in data extraction were resolved by discussion, with adjudication by a third reviewer when necessary. For each prespecified outcome, we extracted data at baseline and at the end of the intervention (postintervention values and/or change scores, as reported), prioritizing the end-of-intervention assessment for the primary synthesis. Risk-of-bias information for each included study was extracted in parallel to support risk-of-bias assessments and GRADE judgments. When required data were missing or unclear, attempts were made to contact study authors; otherwise, available data were analyzed as reported. No automation tools were used for data extraction. As only English-language reports were eligible, no translation was required for data collection.

Outcome data were extracted for the prespecified outcome domains and measurement instruments, with scale directions checked to ensure consistent interpretation. Outcomes of interest included:

For each outcome, we extracted the assessment time points and baseline and end-of-intervention data (postintervention values and/or change scores as reported), including the mean and SD or SE and sample size per group.

The following additional data items were extracted:

Risk of bias was assessed at the study level for included RCTs using the Cochrane Risk of Bias tool (RoB 1), following the Cochrane Handbook (version 5.1.0). The following domains were evaluated, including random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias. Each domain was judged as low, high, or unclear risk of bias. An overall risk-of-bias judgment was derived as follows: studies were rated low overall if all domains were low risk, high overall if one or more domains were high risk, and unclear overall otherwise. Two reviewers (JW and WZ) independently performed the assessments, and disagreements were resolved through discussion, with adjudication by a third reviewer (QZ) when necessary. Where required, study authors were contacted to clarify information relevant to risk-of-bias judgments. No automation tools were used for risk-of-bias assessment.

All prespecified outcomes were continuous measures. For outcomes assessed using the same instrument and reported on the same scale across studies, we synthesized effects as mean differences (MDs) with 95% CIs (eg, 10MWT time [seconds], stride length [meters], double support time [% gait cycle], UPDRS III [points], EQ-5D utility, FES-I [points]). When the same construct was assessed using different instruments, we used standardized mean differences (SMDs; Hedges g) with 95% CIs (eg, pooling balance outcomes across BBS, Mini-BESTest, and POMA balance; pooling quality-of-life outcomes across PDQ-39 and PDQ-8). Effect directions were checked before synthesis to ensure consistent interpretation across scales (eg, lower scores indicating improvement for PDQ, FOG-Q and NFOG-Q, UPDRS III, and 10MWT time; higher scores indicating improvement for balance scales).

For each outcome, we prioritized end-of-intervention assessments. We preferentially extracted post-intervention values; when only change-from-baseline data were reported, these were used as reported. Where change scores were not reported but could be derived from available pre- and postintervention summary statistics, they were calculated using standard formulas; otherwise, postintervention values were used. For all meta-analyses, effect estimates were presented with 95% CIs and, when 3 or more studies were available, 95% prediction intervals. We did not prespecify universal thresholds for “small, moderate, or large” effects; interpretation emphasized clinical relevance on the original scale (for MDs) and certainty of evidence (GRADE).

A total of 2300 relevant articles were identified. After the initial screening, 866 duplicate articles were excluded. Based on the titles and abstracts, a further 1394 articles were excluded. Following a full-text review according to the inclusion and exclusion criteria, 31 additional articles were excluded. Ultimately, 9 studies were included in the analysis [25,40-47]. The literature screening process is shown in Figure 1. Full-text reports that appeared potentially eligible but were excluded, together with reasons for exclusion, are listed in S2 in Multimedia Appendix 1.

Figures 2A and 2B summarize the risk-of-bias assessment [25,40-47]. Overall, methodological quality was mixed. Most trials reported adequate random sequence generation, whereas details of allocation concealment were frequently insufficient or not reported, leading to many domains being judged as having an unclear risk of bias. Because blinding of participants and personnel was largely not feasible given the nature of wearable-device interventions, several studies were rated at high risk of performance bias. In contrast, blinding of outcome assessors was implemented in approximately half of the trials. Incomplete outcome data and selective reporting were generally judged to be at low risk of bias, with most studies reporting attrition and prespecified outcomes adequately. Other potential sources of bias were usually rated as unclear.

A total of 9 studies were included [25,40-47], all of which were published in English, involving a total of 260 PD patients. In most studies, the average age of patients was over 60 years, and the average disease duration exceeded 3 years. Two studies [42,45] conducted 10MWT analysis, 4 studies [40,41,43,46] analyzed stride length, 3 studies [41,43,46] analyzed DST, 4 studies [41,42,44,45] analyzed FOG-Q, 5 studies [41,42,45-47] analyzed UPDRS III, 5 studies [41,42,45-47] analyzed balance function, 3 studies [42,45,47] analyzed PDQ-39, 2 studies [25,45] analyzed EQ-5D utility, and 2 studies [41,44] analyzed FES-I. In most of these studies, patients maintained stable medication regimens before and during the treatment period. Detailed information on the patients and wearable device variables used in these studies is shown in Table 2.

According to GRADE, the overall certainty of evidence ranged from low to very low for all outcomes. For stride length, the certainty was rated as low, reflecting consistent direction of effect across four randomized trials and reasonably precise estimates, but downgrading for very serious risk of bias related to unclear allocation concealment, lack of blinding of outcome assessors, and incomplete outcome data. For all other prespecified outcomes—including 10MWT, DST, freezing of gait, UPDRS III, balance scales, PDQ, EQ-5D utility, and FES-I—the certainty of evidence was judged very low. These ratings were primarily driven by very serious risk of bias and very serious imprecision, with small sample sizes per outcome and wide confidence and prediction intervals that encompassed both no effect and potentially important benefit or harm. Inconsistency and indirectness were generally not considered serious (Table 3). Because the number of included studies was small, we could not formally evaluate funnel plot asymmetry; therefore, the possibility of small-study effects could not be ruled out [48].

This systematic review and meta-analysis synthesized RCTs evaluating the effects of wearable-device interventions on gait, balance, and quality of life in people with PD, interpreting the findings within the ICF framework [49]. We found a small but statistically significant improvement on average in stride length; however, most other outcomes—including short-distance gait speed, double support time, freezing of gait, balance scales, health-related quality of life, and fear of falling—did not show clear between-group differences. These results suggest that current wearable technologies may have only modest effects on specific gait parameters at the level of body functions and activities, with insufficient evidence for consistent improvement in balance or overall quality of life. Importantly, prediction intervals for several outcomes crossed the null, indicating that effects may vary across populations and implementation contexts and may be minimal or absent in some settings.

Regarding gait and motor outcomes, wearable devices produced a small improvement on average in stride length with low statistical heterogeneity, although the prediction interval suggests that effects may be minimal in some settings, consistent with the idea that cueing and feedback can help increase step amplitude in PD. In contrast, we did not observe clear between-group differences in 10-meter walk performance, double support time, freezing-of-gait scores, or UPDRS III, and the corresponding prediction intervals generally spanned both small benefits and no effect. Overall, these results point to at most modest and uncertain benefits of wearable-device interventions on gait and motor severity. Our findings on short-distance gait performance differ from those of a previous meta-analysis by Zhang et al [30], which reported a statistically significant improvement in gait speed with wearable cueing devices. One likely explanation is that Zhang et al [30] pooled several conceptually different outcomes (10MWT, 6-minute walk test, and treadmill-based speed measures) under a single “gait velocity” construct, whereas we analyzed the 10-Meter Walk Test separately and did not combine it with endurance or treadmill outcomes [30]. In addition, we used a more conservative random-effects approach (REML with Hartung–Knapp adjustment and prediction intervals [36]), which yields wider uncertainty intervals and may partly explain why the apparent speed benefit did not remain statistically significant or robust across settings in our analyses. The small but statistically significant improvement in stride length, in the absence of clear effects on 10MWT or freezing measures, may reflect that step amplitude is more directly modulated by external cueing and feedback than more complex phenomena such as gait initiation, freezing episodes, or sustained walking speed; moreover, many participants were in Hoehn and Yahr stages 1-3 with relatively preserved gait, which may limit the observable incremental benefit of wearable devices on standard clinical tests [50].

For balance-related outcomes, this meta-analysis is, to our knowledge, the first to synthesize randomized evidence on the effects of wearable-device interventions on balance function in people with PD. We pooled BBS, Mini-BESTest, and POMA balance subscores using SMDs, as they reflect a common construct of postural control within the ICF domain of body functions and activities. The combined analysis suggested a borderline improvement in balance favoring wearable devices, but the confidence and prediction intervals were wide and included no effect, indicating substantial uncertainty. A key contextual factor is that, among the included trials, only one study explicitly designed the wearable intervention as a balance-focused training program, whereas the others primarily targeted gait initiation or step regulation; thus, the cueing content and training priorities may not have been optimal for producing measurable changes on balance scales. In contrast, a meta-analysis in a nonspecific older-adult population by Mao et al [51] reported clearer balance gains from sensor-based interventions, suggesting that differences in underlying pathology, intervention content, and training dose between PD cohorts and general older adults may be important.

Quality-of-life and fear-of-falling outcomes map more closely to the ICF domains of participation and contextual or personal factors [49]. Across trials, wearable-device interventions did not result in statistically significant improvements in PDQ scores, EQ-5D utility, or FES-I scores, and, where calculated, prediction intervals were wide or included no effect. This pattern suggests that short-term, device-focused interventions may not be sufficient to translate modest improvements in gait or motor scores into perceived gains in overall quality of life or reduced fear of falling. Nonmotor symptoms, psychosocial and environmental factors, and real-world participation constraints were rarely targeted explicitly in the included trials, which may further limit the impact of wearable devices on these broader patient-centered outcomes. From a clinical perspective, these findings underscore that wearable devices are unlikely to replace comprehensive multidisciplinary rehabilitation; rather, they may serve as adjunct tools within a broader program that also addresses cognition, mood, balance confidence, and environmental adaptation [52].

When the findings were interpreted in light of heterogeneity, risk of bias, and GRADE assessments, the overall certainty of evidence for most outcomes was low to very low. Although we restricted inclusion to RCTs, several studies had unclear or high risk of bias in domains such as allocation concealment, blinding of outcome assessment, and incomplete outcome data. Many meta-analyses included only 2-5 trials, leading to wide confidence and, where applicable, prediction intervals that encompassed both no effect and potentially clinically relevant benefit or harm. This imprecision, together with some inconsistency in results and indirectness arising from heterogeneous devices and protocols, resulted in downgrading of GRADE ratings for multiple outcomes [39]. Consequently, our conclusions should be interpreted with caution and considered hypothesis-generating rather than definitive.

This study has several limitations that should be considered when interpreting the results. First, the number of eligible trials and the total sample size for each outcome were modest, limiting statistical power and precision, especially for balance, quality-of-life, and fear-of-falling measures. Second, there was substantial diversity in the technical characteristics and application methods of the wearable devices (eg, cueing modality, feedback algorithms, and sensor placement) and in training protocols (setting, frequency, and duration), which may have diluted or masked device-specific effects. Third, most participants were in Hoehn and Yahr stages 1-3, with relatively preserved motor function and independence, restricting the generalizability of our findings to individuals with more advanced disease. Fourth, intervention periods and follow-up durations were relatively short, so the long-term sustainability of any observed benefits remains unknown. Finally, the number of studies per outcome was too small to formally evaluate funnel plot asymmetry; therefore, small-study effects cannot be ruled out.

Several limitations of our review processes should be acknowledged. First, we restricted eligibility to English-language reports, which may have introduced language bias and could have led to missed trials published in other languages. Second, our search did not include gray literature sources (eg, preprint servers, theses, conference proceedings, or general search engines), and we did not perform forward citation tracking; consequently, relevant unpublished or difficult-to-index studies may not have been identified. Third, because all outcomes were informed by fewer than 10 studies, formal assessments of small-study effects (eg, funnel plots and Egger test) were not feasible; therefore, reporting bias due to missing results cannot be excluded. Finally, although we contacted study authors when key information was unclear or missing, incomplete responses may have limited data availability for some syntheses, potentially contributing to imprecision.

This review provides an updated and methodologically rigorous synthesis of RCT evidence on wearable devices for gait, balance, and quality of life in Parkinson disease. Innovatively, we integrated the ICF framework with contemporary random-effects meta-analysis (Hartung–Knapp adjustment and prediction intervals) and GRADE to contextualize both the pooled average effects and the expected range of effects across real-world settings. Compared with prior reviews, we avoided pooling conceptually different gait outcomes into a single construct, enabling a more implementation-relevant interpretation of heterogeneity and generalizability. This work contributes to the field by clarifying where evidence is most consistent (eg, stride length) and where effects remain uncertain across settings, particularly for balance and patient-centered outcomes. In real-world practice, wearable technologies are best viewed as adjuncts rather than stand-alone rehabilitation strategies and may need to be integrated with task-specific gait and balance training, behavioral strategies, and environmental modifications to translate modest gait gains into meaningful improvements in confidence and quality of life. Future trials should enroll larger and more diverse populations, extend intervention and follow-up periods, standardize ICF-aligned outcomes, and transparently report device characteristics and implementation strategies to clarify for whom, under what conditions, and how wearables can be optimally integrated into routine care.

This systematic review and meta-analysis of randomized controlled trials suggests that wearable-device interventions may provide modest benefits on average for specific gait parameters in PD, particularly stride length, but do not consistently improve short-distance gait speed, double support time, freezing of gait, balance performance, health-related quality of life, or fear of falling. Within the ICF framework, current evidence indicates that wearable devices may support selected body functions and activities, yet there is insufficient and uncertain evidence that they reliably enhance broader participation or overall quality of life.

When interpreted alongside heterogeneity estimates, prediction intervals, risk-of-bias assessments, and GRADE ratings, the certainty of evidence for most outcomes is low to very low. Many meta-analyses were based on a small number of heterogeneous trials with modest sample sizes and relatively short intervention and follow-up durations, leading to wide ranges of plausible effects across settings, including minimal or no benefit. Accordingly, our findings should be viewed as cautious and hypothesis-generating rather than definitive and do not currently justify the use of wearable devices as a stand-alone rehabilitation strategy in PD.

From a clinical perspective, wearable technologies are best considered as adjuncts to comprehensive multidisciplinary rehabilitation. Combining wearable cueing and feedback with task-specific gait and balance training, education, and psychosocial and environmental interventions may be required to translate small gains in gait parameters into meaningful improvements in balance confidence, participation, and quality of life. Future trials should recruit larger and more diverse patient populations, including those with more advanced disease; standardize outcome measures across ICF domains; extend intervention and follow-up periods; and transparently report device characteristics and implementation strategies to clarify which patients benefit most and how wearable devices can be optimally integrated into routine care.

No external financial support or grants were received from any public, commercial, or not-for-profit entities for the research, authorship, or publication of this article.

The analytic code used for the meta-analyses and figure generation and the extracted data underlying the analyses are available from the corresponding author upon reasonable request. The data collection forms and additional materials are available upon request.