Effectiveness of Blood Flow Restriction Training in Neurological Rehabilitation: A Systematic Review and Meta-analysis

Abstract

Introduction

Muscle weakness is a common impairment following neurological disorders, yet traditional high-load training aimed at addressing this is often impractical or unsafe for these patients. Blood flow restriction (BFR) training has emerged as a viable method, capable of eliciting physiological adaptations comparable to high-load training while using significantly lower mechanical loads.

Objective

This study aimed to summarize the clinical applications of BFR training for neurological disorders and assess its effectiveness and safety in improving muscle strength and functional outcomes.

Methods

The systematic review and meta-analysis followed PRISMA guidelines. PubMed, Embase, PEDro, Cochrane, CINAHL, CNKI were searched for intervention studies comparing BFR with diverse control interventions. The primary outcomes were muscle strength and functional outcomes. Sensitivity analyses, subgroup analysis and meta-regression were also performed.

Results

Thirty-seven studies involving 963 participants were included, with individuals having stroke (n = 639), Parkinson's disease (n = 18), multiple sclerosis (n = 128), spinal cord injury (n = 170), cerebral palsy (n = 1), peripheral nerve injury (n = 2), and transverse myelitis (n = 1). Our results revealed that the addition of BFR significantly increased muscle strength, with effect sizes of 0.752 (95% CI 0.49–1.02). Significant improvements were also observed in balance, Fugl-Meyer Lower Extremity assessment, and Modified Barthel Index. Subgroup analyses revealed that onset time was a crucial factor for the effectiveness of BFR on clinical outcomes. Furthermore, BFR training demonstrated overall safety in neurological rehabilitation.

Conclusion

These findings support the effectiveness of BFR training in improving muscle strength and functional outcomes in neurological rehabilitation, without any significant adverse events. However, further high-quality research is needed to validate the efficacy of BFR and refine its training protocols for individuals with neurological disorders.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40798-026-01022-z.

Key Points

1. BFR training, when added to conventional therapy, significantly improves muscle strength and key functional outcomes (including balance, Fugl-Meyer Lower Extremity assessment, and Modified Barthel index) in patients with neurological disorders.

2. The benefits of BFR training on functional outcomes are most significant when initiated early (within 3 months of onset) and when the intervention duration is longer than 4 weeks.

3. BFR training is well-tolerated in neurological populations, with no serious adverse events reported when appropriate screening and monitoring protocols are implemented.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40798-026-01022-z.

Introduction

Neurological disorders are now a major and increasing global health challenge, affecting millions of people across all ages [1]. According to the Global Burden of Disease Study 2016, neurological disorders comprise 11.6% of global disability-adjusted life-years for all diseases [2]. Common neurological disorders include stroke, Parkinson’s disease, spinal cord injuries, multiple sclerosis, cerebral palsy, and peripheral nerve injuries. These conditions often result in varying degrees of motor disability [3] that can ultimately impede the performance of daily activities and pose a great threat to quality of life [4].

Rehabilitation therapy plays an indispensable role in the maintenance and recovery of motor function in individuals with neurological disorders. The UK National Clinical Guideline for Stroke (2023) recommends at least 3 h of daily motor therapy for stroke patients seeking motor recovery [5]. A clinical practice guideline (2024) on the management of acute spinal cord injury points out that intensive rehabilitation for 12–18 months is required after injury [6]. So far, various treatment techniques for neurological disorder rehabilitation have been continuously evolving [7]. However, the complex nature of neurological disorders makes subsequent rehabilitation resemble the proverbial “black box” [1, 8, 9], thereby posing challenges in discerning key contributors and attributing functional improvements post-treatment. Evidence-based clinical practice is a powerful tool for opening the “black box”, as it allows us to retain effective and efficient treatment methods while eliminating therapies that lack evidence or have been disproven.

Muscle weakness is a common sequela of neurological disorders [10], presenting with manifestations such as muscle atrophy or wasting [11], susceptibility to exercise-induced fatigue [12], impaired balance [13], and increased risk of falls [14]. Muscle weakness can occur in various neurological disorders, including stroke [15], cerebral palsy [16], spinal cord injury [17], multiple sclerosis [18], and Parkinson's disease [19]. It can manifest either as an initial symptom or as a result of complications or factors during the progression of the disease. Accumulating evidence suggests that strength training is a useful therapy to address weakness in individuals with neurological disorders [20].

However, the traditionally recommended resistance training loads exceeding 70% 1RM for muscle strengthening [21] may not be practical or safe for patients with neurological disorders, given clinical factors such as limb paralysis, spasticity, and limited tolerance to moderate or high-load exercise. Furthermore, a growing body of evidence indicates that low-load resistance training can elicit comparable levels of muscle hypertrophy when muscle contractions are performed to task failure [22]. In summary, sufficient fatigue-induced muscle activation appears to be a relevant factor for muscle hypertrophy, as low loads (20%, 30%, or 40% 1RM) produce similar hypertrophic outcomes to high loads (80% 1RM) when performed close to failure—a finding consistent with several previous studies [23].

The combination of low-load resistance training with blood flow restriction (BFR) training has emerged as a viable method for managing muscle weakness in neurological disorders. BFR is characterized by the partial restriction of arterial inflow and substantial occlusion of venous outflow in the exercising musculature through the application of a tourniquet to the proximal portion of a limb [24, 25]. This hemodynamic modulation leads to localized venous pooling distal to the occlusion site [24], which accelerates metabolic stress. Physiologically, these mechanisms may elicit muscle mass gains comparable to those from traditional high-load resistance training (HLRT) despite using significantly lower mechanical loads [26, 27]. Consequently, similar levels of muscle adaptation can be attained at reduced external exercise intensities [28]. These hemodynamic responses are transient and are likely attenuated by the muscle pump effect during dynamic contractions [29], making BFR a safe therapeutic adjunct for muscle training with minimal long-term cardiovascular consequences [30].

In the past decade, there has been a growing interest in using BFR for rehabilitating neurological diseases. The application has gained global adoption across Asia, North America, Europe, South America and the Middle East [31]. The potential therapeutic value of BFR in neurorehabilitation is highly encouraging and comparable to its successful use in treating musculoskeletal diseases [32]. In a systematic review published in 2022, the effects of BFR training on neurological disorders were summarized across seven articles covering four diseases. The findings demonstrated the safety and promising prospects of BFR training [33]. The included studies showed improvements in sensorimotor function, gait symmetry and speed, perceived exertion, walking endurance, quality of life, muscle thickness and density, and a reduction in muscle edema. However, no improvements were observed in balance. It is important to note that caution should be exercised when interpreting these results due to the limited number of articles included and the heterogeneity of participants and training parameters.

With the growing number of BFR studies emerging in neurological rehabilitation, a quantitative meta-analysis can now be conducted to provide more definitive evidence and analyze possible sources of heterogeneity. Before widely implementing BFR training in this population, it is imperative to understand the physiological responses and physical adaptations compared with conventional therapy. Additionally, determining optimal training parameters based on current evidence is important for promoting this novel technique and gaining a new perspective on neurological rehabilitation. To our knowledge, there are no existing systematic reviews that provide quantitative synthetic evidence on this topic.

Therefore, the aim of this systematic review and meta-analysis is to evaluate the effectiveness of BFR training in neurological disorders and provide a current, research-informed guide to facilitate further clinical application.

Methods

The systematic review was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO, CRD42023428002) and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses guidelines [34].

Search Strategy

A literature search was conducted in the following databases (PubMed, Embase, Cochrane, CINAHL, PEDro, China National Knowledge Infrastructure) from database inception to 1st October 2025. The search strategy is presented in Supplementary Material Table S1-S6. Additional articles were included by screening reference lists from other published reviews on similar topics. A search revision was also scheduled while finalizing the manuscript to avoid missing recently published articles.

Inclusion and Exclusion Criteria

After combining retrieved articles from different databases, the Rayyan tool (https://www.rayyan.ai/) [35] was used to identify and remove duplicates. Two reviewers then independently screened titles and abstracts via the platform. Studies deemed irrelevant at this stage were excluded. Subsequently, the full texts of potentially eligible studies were retrieved and critically appraised to determine final inclusion. Where critical methodological or outcome-related information was missing from the published reports, the corresponding authors were contacted via email to request supplementary data.

The PICOS framework was adopted to define the inclusion and exclusion criteria of qualitative analysis. Studies were included based on the following criteria: (1) population: patients with neurological disorders as defined by the World Health Organization [36]; over 18 years of age; (2) intervention: any type of BFR training, including BFR alone or in combination with conventional therapy; any reliable BFR equipment (BFR cuff, surgical tourniquet, elastic cuff, etc.) applied to the working muscle; (3) comparison: any comparator, including active controls (e.g., conventional resistance training without BFR), sham controls (e.g., placebo BFR with minimal pressure), or no controls; (4) outcomes: muscle strength, muscle size and functional outcome measures; (5) study design: interventional studies; (6) language: English and Chinese. Studies were excluded based on the following criteria: (1) full text unavailable; (2) unable to obtain complete information; (3) protocols, conference articles, animal experiments or reviews; (4) studies with duplicate samples measuring the same outcomes. Studies that met the inclusion criteria for the systematic review but were designed as case reports, case series, or single-subject designs were excluded from the quantitative synthesis. Furthermore, studies were excluded from the quantitative synthesis if they lacked sufficient statistical information (such as means and standard deviations) for any of the comparable outcomes. These criteria were applied to minimize heterogeneity and maintain the methodological rigor necessary for valid pooling of effect sizes.

Two reviewers (YFC and LYC) performed the process of study selection and inclusion according to the search strategy. Any disagreements were resolved through discussion, with a third reviewer (HWW) involved in making the final decision in case of disputes.

Data Extraction

Two independent reviewers (YFC and LYC) extracted data from the selected studies using a standardized extraction custom spreadsheet formed for the study. The extracted data were cross-checked to confirm accuracy. Descriptive data mainly included basic information (title, first author, year of publication, neurological disease or pathology, patient characteristics etc.), study design (cuff size, cuff pressure, training duration, frequency, load) and results obtained (outcomes, adverse events, measurement tools). We extracted baseline data recorded before intervention and outcomes were measured as soon as possible after the treatment and excluded other time points.

Quality Appraisal

Multiple quality assessment tools were employed, depending on the specific study design used. The PEDro Scale was used to assess the methodological quality of the included randomized controlled/crossover trials, and studies scoring below 4 points were excluded [37]. Single-subject research designs were rated using Recommendations for Levels of Evidence and Quality Rating for Single-subject Research Design [38]. The risk of bias in included studies was evaluated using the Cochrane Collaboration Tool [39]. Two reviewers (YFC, LYC) assessed the methodological quality and risk of bias. The results were cross-checked, and any discrepancies were evaluated by a third reviewer (HWW) to reach a consensus.

Synthesis of Results and Statistical Analysis

Calculation of Effect Sizes from Raw Data

Continuous variables were extracted as means, standard deviations (SDs), and sample sizes for both the experimental and control groups before and after the intervention.

When change-from-baseline data were not directly reported, mean changes and their corresponding SDs were calculated independently for each group. The mean change was calculated as:

The SD_change was estimated to account for the within-group correlation (Corr) between pre- and post-intervention measurements. As this correlation coefficient was typically not reported, it was assumed to be 0.5, a conventional estimate recommended by the Cochrane Handbook [39]. The calculation was as follows:

All analyses were conducted in R (version 4.5.1). Effect sizes were computed with the metafor package (v4.8–0) and clubSandwich (v0.6.1) packages. The primary effect-size metric was the standardized mean difference (SMD) corrected for small-sample bias (Hedges’ g). Using the escalc() function, Hedges’ g and its sampling variance were calculated from the study-level change scores (Mean_change, SD_change, and n) for both groups.

Meta-analytic Modeling with Robust Variance Estimation

Many studies reported multiple outcome measures from the same sample, making effect sizes statistically dependent. To appropriately account for this nested data structure, we fitted a three-level random-effects model using rma.mv() function. The model structure was specified as random =  ~ 1 | study_id/effect_id, which nested effect sizes (Level 2) within independent studies (Level 3). This partitions the total variance into three components: sampling variance (Level 1), within-study (between-effect-size) variance (, Level 2), and between-study variance (, Level 3).

Because some analyses included a relatively small number of studies, we applied robust variance estimation (RVE) using the clubSandwich package, with the CR2 small-sample correction and Satterthwaite degrees of freedom to obtain robust tests and confidence intervals. Unless otherwise stated, the pooled effect sizes, confidence intervals, and p-values reported in the main text refer to the RVE results.

Heterogeneity and Publication Bias

Sources of heterogeneity were summarized by the variance components from the three-level model: the within-study variance (, Level 2) and the between-study variance (, Level 3). In addition, to report conventional heterogeneity statistics (Cochran’s and ), we first aggregated multiple effect sizes within studies using inverse-variance fixed-effect weighting to obtain one effect per study and then fitted a standard random-effects model with rma.uni() to the aggregated dataset. The resulting overall and the -test p-value are reported; was interpreted as high heterogeneity.

We performed a leave-one-study-out analysis at the study (cluster) level: the three-level model was re-estimated iteratively after removing one study at a time to evaluate the influence of each study on the pooled effect.

Publication bias and small-study effects were examined on the aggregated data using metafor: visual funnel plots and Egger’s regression test when there were at least 10 single outcome indicators. When Egger’s test indicated significant asymmetry (p < 0.05), we applied the trim-and-fill procedure to estimate potentially missing studies and the implied impact on the pooled effect.

Subgroup Analyses

To explore potential sources of heterogeneity, we conducted prespecified subgroup analyses using mixed-effects models (rma.mv). We tested for differences between categorical subgroups based on key study-level factors, including onset time, duration of intervention, type of neurological disorder, and language of publication.

Results

Study Selection and Methodological Quality

The PRISMA diagram summarized the study selection process (Fig. 1). Out of the initial 3075 studies retrieved from the database, 57 were screened for eligibility in full text. A total of 37 studies met the criteria for systematic review. Among these studies, 23 were randomized controlled trials, 8 were case reports, 1 was a case series, 2 were cohort studies, 1 was a retrospective study, 2 followed single-subject designs, and 21 were included in the quantitative synthesis.

Fig. 1.

PRISMA 2009 flow diagram of study selection process

The mean PEDro score of the included randomized controlled trials was 6.13/10 points (range 5–8) (Supplementary Material Table S7), indicating overall good quality. The two single-subject research design studies scored 9/14 and 10/14, respectively (Supplementary Material Table S8), indicating moderate quality.

Study Characteristics and BFR Training Interventions

A total of 963 patients were included: stroke [11, 40–52] (n = 639), Parkinson’s disease [53–55] (n = 18), multiple sclerosis [13, 56–60] (n = 128), spinal cord injury [61–67] (n = 170), cerebral palsy [59, 68, 69] (n = 5), peripheral nerve injury [70, 71] (n = 2) and transverse myelitis [72] (n = 1). All studies were published between 2014 and 2025. An overview of the characteristics of the included studies is presented in Table 1.

Table 1.

Characteristics of the studies included in the qualitative analysis

Study		Population					BFR group					Control group
References	Type	Total(BFR/CG)	Onset time: BFR/CG	Sex: BFR/CG	Age ± SD: BFR/CG	Duration of intervention; frequency; duration of single training	Device	Location	Cuff width	Pressure, mmHg	Intensity	Intervention	Intensity
Stroke
Ahmed et al. [11]	RCT	30 (15/15)	25.6 ± 29.89/23.06 ± 31.45 (mth)	10 M/9 M	54.2 ± 13.19/ 54.73 ± 14.08	5 wks; 3 times/wk; 40 min	Tourniquet	Affected proximal thigh	not reported	150 mmHg-160 mmHg	10 reps/set, 3 sets	HLRT + AE	80% of 1-RM
Peng et al. [50]	RCT	47 (23/24)	39.48 ± 11.29/ 37.13 ± 11.09 (d)	16 M/14 M	57.7 ± 7.58/ 57.28 ± 7.00	4 wks; 5 times/wk; 20 min	Theratools	Under armpit on the affected side	7.5 cm	60% SBP	Not reported	tDCS	2.0 mA; 20-40 Hz
Du et al. [44]	RCT	60 (30/30)	Not reported	17 M/18 M	52.1 ± 3.9/ 51.8 ± 4.2	6 wks; 5 times/wk; 20 min	B Strong pressure bands (B Strong, Stray Whales, LLC, USA)	Affected mid-femur	Not reported	250 mmHg	10 min/set, 2 sets	TMS	20 Hz; 20 pulses per sequence, 100 sequences
Yang [47]	RCT	66 (33/33)	17.95 ± 3.52/ 18.23 ± 3.62 (d)	15 M/14 M	46.26 ± 5.21/46.26 ± 5.21	6 wks; 5 times/wk; 20 min	Not reported	Affected upper limb, thigh, mid-femur	7 cm	Upper limb: 60-80 mmHg; Thigh: 80-100 mmHg; Mid-femur: 160-170 mmHg	10 min/set, 2 sets	CRT	10 min/set, 2 sets
Tang [46]	RCT	32 (16/16)	32.38 ± 12.31/26.56 ± 14.26 (d)	13 M/14 M	51.50 ± 12.01/52.50 ± 9.03	2 wks; 5 times/wk; 30 min	Not reported	Affected proximal thigh	Not reported	70 mmHg	30–15-15–15, 4 sets, 75 reps	CRT	30–15-15–15, 4 sets, 75 reps
Feng et al. [42]	RCT	29 (15/14)	12.27 ± 11.55/ 14.36 ± 15.45 (mth)	10 M/11 M	54.07 ± 10.44/ 45.29 ± 13.74	3 wk; 7 times/wk; 20 min	B Strong pressure bands (B Strong, Stray Whales, LLC, USA)	Affected proximal thigh	Not reported	160-200 mmHg	20 min/set, 1 set	CRT	30 min/set, 2 sets
Zhang [48]	RCT	60 (30/30)	28.3 ± 12.97/29 ± 11.93 (d)	22 M/19 M	53.87 ± 10.69/54.70 ± 11.51	6 wks; 5 times/wk; 20 min	Tourniquet	Affected mid-femur	7 cm	160-170 mmHg	10 min/set, 2 sets	NMES	30–45 Hz, 200-300 μs, 6–20 mA
Du et al. [43]	RCT	60 (30/30)	76.4 ± 18.4/75.6 ± 11.1 (d)	18 M/17 M	66.8 ± 11.2/67.2 ± 10.8	8 wks; 5 times/wk; 30 min	Tourniquet	Affected mid-femur	7 cm	140-200 mmHg, 1.3 SBP	10 × 20 reps, 10–30% 1RM	CRT	10 × 20 reps, 10–30% 1RM
Chen and Yu [40]	RCT	71 (35/36)	32.3 ± 5.5/31.6 ± 6.4 (d)	22 M/21 M	65.8 ± 10.2/66.9 ± 11.7	12 wks; 5 times/wk; 30 min	Tourniquet	Affected mid-upper arm	7 cm	140-200 mmHg, 1.3 SBP	10 × 20 reps, 10–30% 1RM	CRT	10 × 20 reps, 10–30% 1RM
Xu [45]	RCT	58 (29/29)	1.58 ± 1.28/1. 6 ± 1.27 (mth)	19 M/24 M	60.1 ± 11.2/59.41 ± 11.85	2 wks; 12 times/wk; 20 min	Tourniquet	Upper margin of patella 10 cm on the affected side	10 cm	200-240 mmHg, 60% AOP	3 min/set, 5 set	CRT + AE	3 min/set, 5 set, 20% 1RM
Yue [41]	RCT	26 (13/13)	27.00 ± 7.83/28.08 ± 7.18 (d)	9 M/10 M	51.62 ± 4.48/53.15 ± 4.65	6 wks; 12 times/wk; 30 min	Tourniquet	Upper thigh	Not reported	160-180 mmHg	30–15-15–15, 30% 1RM	CRT	30–15-15–15, 30% 1RM
Kjeldsen et al. [49]	RCD	28 (14/14)	241.93 ± 167.30/ 241.93 ± 167.30 (d)	7 M/7 M	58.43 ± 9.48/58.43 ± 9.48	2 apart days; 2 times, spaced 7 ± 1 days; Not reported	Reister®	Thigh	22/26 cm	94-120 mmHg, 80%SBP	30–15, 2 sets, 45 reps	CRT	30–15, 2 sets, 45 reps
Li et al. [52]	RCT	34 (17/17)	47.06 ± 42.26/ 41.53 ± 33.50(d)	14 M/15 M	59.59 ± 8.95/ 64.41 ± 12.90	3 wks; 10 times; 20 min	Not reported	Proximal lower limbs near the groin	10 cm	40%-80% AOP	10 min/set, 2 sets	Sham	10 min/set, 2 sets,10% AOP
Feng et al. [51]	RCT	38 (20/18)	7.00(4.00 ~ 23.50)/6.00(3.50 ~ 24.00) (mth)	15 M /15 M	52.70 ± 11.12/ 44.85 ± 12.99	4 wks; 5 times/wk; 30 min	B Strong pressure bands (B Strong, Stray Whales, LLC, USA)	Proximal portion of the thigh	Not reported	160–200 mmHg	30 min	CRT	10 15 reps/set, 30 min/set, 2 sets
Parkinson’s disease
Douris et al. [54]	SSRD	1	7 (yr)	1 M	65	6 wks; 3 times/wk; Not reported	Blood pressure cuff	Upper thigh	Not reported	120-160 mmHg	5 sets, 2 min/set, 50 m/min	–	–
Douris et al. [53]	SSRD	1	8 (yr)	1 M	66	6 wks; 3 times/wk; Not reported	Blood pressure cuff	Upper thigh	Not reported	Began with 50%SBP and increased by 10% each week to a final pressure of 100% of SBP	30% 3RM, 3 × 15reps	–	–
Manago et al. [55]	Prospective cohort	16	7.6 ± 5.7 (yr)	9 M	68.1 ± 8.6	8 wks; 2 times/wk; 60 min	Delfi PTSII	Most proximal portion of the limb	Not reported	Upper limb: 40–50% LOP; Lower limb: 60–80% LOP	30–15-15–15, 20–30% 1RM	–	–
Spinal Cord Injury
Krogh et al. [64]	Case report	1	11 (mth)	1 M	23	4 wks; 2 times/wk; Not reported	Cylindrical Tourniquet Cuff 90 mm, Zimmer Surgical, Inc., Dover, OH, USA	Proximal upper limb	9 cm	100 mmHg	3 sets, 15–25 reps/set, with a 3 kg sandbag	–	–
Skiba et al. [76]	RCT	16 (9/7)	4.89 ± 7.52/ 4.71 ± 5.9 (yr)	9 M/7 M	24.44 ± 6.29/24.57 ± 6.59	8 wks; 2 times/wk; 15 min	Sphygmomanometer and a vascular Doppler device	Around the thigh immediately below the inguinal crease	25 cm	65.2 ± 7 mmHg, 50%AOP	3 sets, 4 min/set	FES	20 Hz, 400 μs wide, 2 to 4 mA
Stavres et al. [62]	RCD	18 (9/9)	> 6 (mth)	9 M/9 M	63 ± 12.15/63 ± 12.15	1 day; 1 time; Not reported	DE Hokanson, Inc, Bellevue, WA	Most proximal region of the least affected leg	Not reported	125% VOP	3 sets, 3 × 10 reps	CRT	3 sets, 3 × 10 reps
Gorgey et al. [63]	Self-control	18 (9/9)	Not reported	9 M/9 M	Not reported	6 wks; 2 times/wk; 30 min	blood pressure cuff	Forearm to occlude the brachial artery	Not reported	111-178 mmHg, 130%SBP	40 contractions	NMES	40 contractions, 20 Hz and 450-μs pulse
Babak et al. [65]	Self-control	20(10/10)	145.6 (mth)	16 M/16 M	47 ± 14.84/47 ± 14.84	8 wks; 2 times/wk; 17 min	Delfi PTS, Delfi Medical, Vancouver, Canada	Exercising arm	Not reported	60% LOP	30,15,15,15	CRT	4 sets of wrist-curl
Jønsson et al. [66]	RCT	21(11/10)	17.3 ± 15.3/11.0 ± 9.5 (yr)	5 M/5 M	58.0 ± 8.3/59.5 ± 9.8	8 wks; 2 times/wk; 45 min	Delfi Medical Innovations Inc. Canada	Horizontally on the thigh near the inguinal fold	11 cm	40% AOP, 75.9 mmHg	30,15,15,15	Sham	30,15,15,15, without inflating the cuff
Xiao et al. [67]	Retrospective study	76(39/37)	6.70 ± 2.35/6.88 ± 2.40(mth)	22 M/22 M	40.85 ± 6.11/41.30 ± 6.24	12 wks; 3 times/wk; 30-50 min	Delfi PTS Personalized Tourniquet System, Delfi Medical Innovations Inc., Vancouver, Canada	Proximally on both thighs	11 cm	50%-60% SBP	20–30% 1RM; 4 sets × 15reps	CRT	12 wks; 3 times/wk; 60-90 min CRT
Multiple Sclerosis
Freitas et al. [56]	RCD	15	Not reported	4 M	45.7 ± 9.4	2 apart days; spaced 14 days; 30 min	E20 Rapid Cuff Inflator, D.E. Hokanson, Bellevue, WA, USA	Most proximal portion of both legs	13.5 cm wide	50% tAOP	30/15/15/15, 4 sets, 20% 1RM	HLRT	30/15/15/15, 4 sets, 70% 1RM
Lamberti et al. [77]	RCT	22 (11/11)	14 ± 9/13 ± 10 (yr)	4 M/3 M	54 ± 11/56 ± 10	6 wks; 2 times/wk; 40 min	The Occlusion Cuff LTD, Somerset, UK	Most proximal portion of both legs	6 cm wide	30% SBP	5 bouts of walking at speed 60 steps/minute	Conventional intensive overground walking	5 bouts of walking at speed 60 steps/minute
Darvishi et al. [57]	RCT	40 (10/10/10/10)	Not reported	5 M/5 M/5 M/5 M	45.31 ± 11.4/43.6 ± 9.3/48.42 ± 2.6/51.11 ± 6.4	8 wks; 2 times/wk; 45 min	Manual pressure barrel	Both proximal thighs	8 cm wide	150-160 mmHg	50- 60% maximum heart rate training	CG1: AE + CRT; CG2: BFR + CRT; CG3: CRT	50- 60% maximum heart rate training
Brown et al. [58]	Case report	1	> 1 (yr)	0 M	30	8 wks; 2 times/wk; 45–60 min	Delfi Personalized Tourniquet System	Most proximal portion of the weaker thigh	Not reported	Increased from 70 to 80% pressure occlusion in 5 weeks	30/15/15/15	–	–
Cohen et al. (2021) [59]	Case report	1	13 (yr)	0 M	54	12 wks; 2 times/wk; 45–60 min	Delfi Medical Innovations, Vancouver, British Columbia, Canada	–	5 inches	40–80% AOP	30/15/15/15, 20–40% 1RM	–	–
Mañago et al. [60]	Cohort Study	14	19.1 ± 10.7 (yr)	4 M	55.4 ± 6.2	8 wks; 2 times/wk; 45–60 min	Delfi PTSII; Delfi Medical Innovations, Vancouver, British Columbia, Canada	Most proximal portion of the thigh	Not reported	60%–80% LOP	30/15/15/15	–	–
Schmidt et al. [74]	RCT	17(8/9)	≥ 6 (mth)	1 M/1 M	39 ± 10/42 ± 13	12 wks; 2 times/wk; 45–60 min	Suji Sub Inc, leAD Sports Office, Tavis tock Lakes Blvd., FL USA	Most proximal portion of the upper arms and legs	4–4.5 inches	60% LOP	30/15/15/15, 30% 1RM	HLRT	4 sets, 8–12 reps/set, 65% 1RM
Cerebral Palsy
Salvador et al. [68]	Case report	1	19 (yr)	1 M	20	4 wks; 3 times/wk; 20 min	Not reported	Bilateral cuff inflation on the upper thighs	18 cm wide	140-170 mmHg	Four bouts of 5 min treadmill walk training at 5.2 km/h	–	–
Cohen et al. [59]	Case report	1	13 (yr)	0 M	54	12 wks; 2 times/wk; Not reported	Delfi Medical innovations, Vancouver, British Columbia, Canada	Thigh	5-inch-wide	80% AOP	30/15/15/15	–	–
Joyce et al. [69]	Case series	3	–	3 M	23, 27, 34	8 wks; 2 times/wk; Not reported SmartTools Plus	Cleveland, Ohio	Most proximal portion of both thighs	10.16 cm	60–80% LOP, 20%1RM	30/15/15/15	–	–
Peripheral nerve injury
Yasuda et al. [70]	Case report	1	Not reported	0 M	43	BFR and CG were performed six times each over multiple days; Not reported; 4-5 min	KAATSU Master, KAATSU JAPAN Co., Ltd., Tokyo, Japan	Most proximal portion of both arms	30 mm width	130–170 mmHg	40–50 dB each performance	Piano control	40–50 dB each performance
Vineyard et al. [71]	Case report	1	Not reported	1 M	23	8 wks; 2 times/wk; Not reported	Kaatsu training, Sato Sports Plaza Inc., Japan	Unaffected upper portion of the humerus, upper thigh	Not reported	Upper limb: 320 mmHg; lower limb: 380 mmHg	3 sets of 30 reps for each exercise, 30% 1RM	–	–
Transverse myelitis
Mintken et al. [72]	Case report	1	17 (yr)	1 M	31	8 wks; 2 times/wk; Not reported	SmartCuffs© PRO; Smart Tools Plus, Strongsville, Ohio	As far proximally as possible on the thigh	Not reported	80% LOP	1 set of 30 repetitions followed by 3 sets of 15 at 20% to 30% of the 1RM and 60% to 80% of the maximal LOP	–	–

RCT, randomized controlled trial; SSRD, Single-subject Research Design; RCD, randomized crossover design; CG, control group; SD, standard deviation; M, male; yr, year; d, day; mth, month; wk, week; wks, weeks; rep, repetition; min, minute; RM, repetition maximum; BFR, blood flow restriction; SBP, systolic blood pressure; CRT, conventional rehabilitation training; tDCS, transcranial direct current stimulation; TMS, transcranial magnetic stimulation; AE, aerobic exercise; HLRT, high load resistance training; cm, centimeter; mA, milliampere; Hz, Hertz; μs, microsecond; LOP, limb occlusion pressure; AOP, Arterial Occlusion Pressure; FES, Functional Electrical Stimulation; NMES, Neuromuscular Electrical Stimulation; VOP, Venous Occlusion Pressure; mmHg, Millimeters of mercury; dB, Decibels.

The data extraction results (Table 1 and Supplementary Material Table S9) indicate variations in the parameters of BFR training across studies. The occlusion pressure ranged from 58.2 to 380 mmHg. The duration of a single training session varied from 4 to 60 min, with a common duration of 20 min. The overall duration of intervention ranged from 1 day to 12 weeks, with three studies focusing on immediate effects [49, 56, 62]. The treatment frequency ranged from twice to seven times per week, with twice per week being the most prevalent. In terms of the intervention setting, twenty-six studies incorporated BFR alongside conventional rehabilitation therapy [11, 40–43, 45–47, 49, 51–53, 55, 56, 58–60, 64–67, 69, 71, 73–75]; four studies combined BFR with electrical stimulation [48, 63, 72, 76]; one study integrated BFR with transcranial magnetic stimulation [44]; one study incorporated BFR with transcranial direct current stimulation [50]; one study supplemented piano performance with BFR [70]; one study added aerobic training to BFR [57]; and three studies combined BFR with walking training [13, 54, 68].

Outcome Measures

Muscle strength was assessed through 3-RM leg press [53], dynamometer [40, 57, 58, 69], manual muscle testing (MMT) [42–44, 48, 50, 51, 69], Lower Extremity Motor Score [67], and isokinetic knee extensor [45]. Motor function outcomes included Fugl-Meyer Lower Extremity assessment (FMA-LE) [42–45, 47, 48, 51], Berg balance scale (BBS) [13, 41, 43, 46, 47, 57, 58], six-minute walk distance [11, 13, 44, 48, 54, 66, 74, 78], five time sit to stand test [11, 13, 74], timed up and go test [11, 42, 51, 53–55, 66], ten-meter walk test [58, 66], timed 25-foot walk test [68], and 30-s chair stand test [53–55, 74]. The modified Barthel index (MBI) [40, 42–44, 48, 50] was utilized to assess activities of daily living (Table 2).

Table 2.

Summary of forest plot results

Outcome	Number of studies	Number of effect sizes	SMD (Hedges' g)	95% CI	Peffect	I2 (%)	Pheterogeneity
1. Muscle strength	12	30	0.752	[0.489, 1.015]	< 0.001	78.16%	0.001
2. Clinical outcome	19	41	0.534	[0.268, 0.799]	0.001	76.61%	< 0.001
FMA-LE	6	6	0.785	[0.287, 1.282]	0.027	76.75%	< 0.001
Berg balance scale	7	7	0.694	[0.170, 1.219]	0.042	73.99%	< 0.001
Gait SPEED	8	11	0.536	[− 0.094, 1.166]	0.139	86.90%	< 0.001
Sit to stand	4	5	0.299	[− 0.057, 0.655]	0.150	0.00%	0.713
Activities of daily living	7	7	0.893	[0.459, 1.328]	0.007	73.01%	0.003

SMD, Standardized Mean Difference; 95% CI, 95% Confidence Interval; p_effect, p-value for the overall effect size; I², I-squared statistic (measure of heterogeneity); p_{heterogeneity}, p-value for the test of heterogeneity; FMA-LE, Fugl-Meyer Assessment for Lower Extremity; Bold text denotes the main outcome categories (or overall pooled estimates), visually distinguishing them from their specific sub-scales

Muscle Strength

Regarding muscle strength, a summary analysis of 12 studies (30 effect sizes) demonstrated a significant, moderate-to-large beneficial effect of BFR intervention (SMD = 0.752, p < 0.001, 95% CI: [0.489, 1.015]). The analysis indicated significant and high heterogeneity (I² = 78.16%, p < 0.001). The three-level model provided a key insight: this heterogeneity was almost entirely attributable to between-study variance (τ₃² = 0.158), with near-zero within-study variance. This finding strongly suggests that the high variability is driven by differences in measurement tools (e.g., MMT vs. dynamometry vs. isokinetic testing), rather than by measuring multiple muscle groups within the same study. Darvishi et al. [57], Joyce et al. [55], Mañago et al. [73] and Jønsson et al. [66] assessed maximum isometric strength of the lower limb using a dynamometer. Chen [40] employed a dynamometer to measure grip strength, while Xu [45] utilized isokinetic muscle strength testing and changes in pedal resistance to evaluate muscle strength in stroke patients. Five studies [42–44, 48, 50] reported MMT outcomes in stroke, and one study [67] reported Lower Extremity Motor Score (LEMS) outcomes. Egger's test found no significant publication bias (p = 0.914). The trim-and-fill method suggested one potentially missing study, but the adjusted SMD (0.722) was highly consistent with the original. The sensitivity analysis confirmed that the findings were highly robust: the pooled SMD remained stable within a range of 0.710 to 0.815 and was highly statistically significant in all scenarios.

Functional Outcomes

FMA-LE

Regarding FMA-LE, data from 6 studies (6 effect sizes) indicated a significant, large positive effect for BFR intervention (SMD = 0.785, p = 0.027, 95% CI: [0.287, 1.282]). High and significant heterogeneity was also observed in this subset (I² = 76.75%, p < 0.001), stemming predominantly from between-study differences (τ₃² ≈ 0.147). This heterogeneity appears to be driven by both the onset time of disease and intervention duration. The primary driver appears to be stroke chronicity. BFR demonstrated a significant, large benefit in patients within the subacute phase [43, 44, 48]. Conversely, it showed no clear superiority over active control therapies in the chronic phase [42, 51]. The study by Xu [45] serves as a key outlier: despite enrolling patients exhibiting a mean disease duration of ≤ 3 months, it reported a minimal effect (SMD = 0.26) similar to that observed in chronic-phase studies. This is likely attributable to the exceptionally short intervention duration of only 2 weeks. While the authors of that study did find objective improvements in muscle strength (isokinetic testing) and activation (sEMG), this brief timeframe was likely insufficient for those underlying gains to translate into significant changes on the FMA-LE functional scale.

No significant evidence of publication bias was found (Egger's test p = 0.651), and the trim-and-fill procedure did not impute any studies. The leave-one-study-out analysis confirmed these findings: removing the Du et al. [44] study, which had the highest effect, caused the pooled SMD to decrease from 0.785 to 0.621.

Balance Function

In the 7 studies (7 effect sizes) assessing balance function (BBS), BFR intervention showed a significant, moderate-to-large positive effect (SMD = 0.694, p = 0.042, 95% CI: [0.170, 1.219]). The analysis revealed high and significant heterogeneity (I² = 73.99%, p < 0.001). Tang [46] implemented a short 2-week intervention and found no significant between-group difference in BBS scores (SMD = 0.22). In contrast, studies by Yue [41] and Yang [47], lasting 6 weeks, demonstrated moderate-to-large positive effects. Yue [41] directly supports this time-dependent effect, reporting no significant difference at 3 weeks but a significant BFR advantage at 6 weeks. This suggests that the benefits of BFR on balance are cumulative and may only become superior to conventional therapy after several weeks. Notable effects were observed in studies targeting specific subgroups: Du et al. [43] reported a large effect (SMD = 2.04) in elderly patients, while Yang [47] found significant benefits (SMD = 0.76) in obese stroke patients. As the number of studies (k = 7) equaled the number of effect sizes (n = 7), the three-level model confirmed that this variance originated entirely between studies (τ₃² ≈ 0.182). The leave-one-study-out analysis revealed the source of this instability: Du et al. [43], who focused on stroke patients using hip muscle training, reported a very large effect (y_i = 2.04). Removing Du et al. [43] caused the pooled SMD to drop from 0.694 to 0.493. Egger's test did not show significant bias (p = 0.426). However, the trim-and-fill method suggested the potential for 2 missing studies, estimating an adjusted SMD that was larger at 0.934 (95% CI: [0.425, 1.443]).

Activities of Daily Living

For activities of daily living, an analysis of 7 studies (7 effect sizes) demonstrated that BFR intervention had a significant, large positive effect (SMD = 0.893, p = 0.007, 95% CI: [0.459, 1.328]). A high degree of heterogeneity was present among these studies (I² = 73.01%, p = 0.003). This heterogeneity appears to be driven by two primary factors: onset time of disease and the characteristics of the control interventions. There is a clear divide based on stroke onset time. Studies focusing on patients exhibiting a mean disease duration of ≤ 3 months [40, 43, 48, 50] reported large positive effects. Conversely, studies recruiting patients in the chronic phase (> 1 year) [11, 42] reported virtually no effect. This pattern strongly suggests that BFR provides a significant additional benefit for ADL during the subacute phase, but in the chronic phase, its efficacy may not be superior to other active therapies. The low effect sizes in the chronic-phase studies are further explained by their use of robust, active comparators. Ahmed et al. [11] compared BFR with low-load training against high-load resistance training; BFR was not found to be superior, resulting in a near-zero SMD. Similarly, the control group in Feng et al. [42] received a double dose of standard exercise training, which likely masked any potential additive benefits from BFR. This division is supported by further statistical analysis. Egger's test on aggregated data approached but did not reach statistical significance (p = 0.063), and the trim-and-fill method did not identify any missing studies. The leave-one-study-out analysis confirmed the findings: removing Ahmed et al. [11], which had almost no effect, caused the pooled SMD to jump to 1.033; removing Du et al. [43], which had a very large effect, caused the SMD to drop to 0.779.

Subgroup Analysis

To explore potential sources of heterogeneity, subgroup analyses were conducted for muscle strength (Table S11) and functional outcomes (Table S12), stratified by onset time, duration, and type of neurological disorder.

For functional outcomes, the analysis by onset time revealed a significant difference between subgroups (p = 0.024). A large and statistically significant effect was observed in the “< 3 months" group (mean difference = 0.75, 95% CI [0.41, 1.10], p < 0.001). In contrast, the effects were small and not statistically significant for the “6 months to1 year" (p = 0.668) and “> 1 year" (p = 0.073) groups. For muscle strength, onset time did not show a significant subgroup difference (p = 0.315).

The analysis by intervention duration (≤ 4 weeks vs. > 4 weeks) revealed no significant subgroup difference for either functional outcomes (p = 0.393) or muscle strength (p = 0.339).

Similarly, the analysis by type of neurological disorder (stroke, multiple sclerosis, and spinal cord injury) also showed no significant differences among subgroups for functional outcomes (p = 0.208) or muscle strength (p = 0.534). This suggests that the type of neurological disorder did not significantly influence BFR's effectiveness, supporting the decision to pool these populations in the main analysis (Fig. 2).

Fig. 2.

Forest plots of the effects of Blood Flow Restriction training on clinical outcomes in patients with neurological disorders. The forest plots illustrate the standardized mean differences and 95% confidence intervals for the following outcome measures: A Effect of BFR on muscle strength: assessing the impact of BFR on various muscle groups and functional strength tasks; B Effect of BFR on FMA-LE: evaluating the recovery of lower extremity motor function using the Fugl-Meyer Assessment; C Effect of BFR on BBS: measuring changes in static and dynamic balance through the Berg Balance Scale; D Effect of BFR on ADL: reflecting the improvement in patients' daily living independence, primarily measured by the Barthel Index or Modified Barthel Index. Individual study effects are represented by squares, with the size of each square proportional to the study's weight in the meta-analysis. The horizontal lines indicate the 95% CIs. The diamond at the bottom of each plot represents the pooled effect size from both fixed-effect and random-effects models. ADL Activities of Daily Living, BBS Berg Balance Scale, BFR Blood Flow Restriction, BI Barthel Index, CI Confidence Interval, FMA-LE, Fugl-Meyer Assessment for Lower Extremity, MBI Modified Barthel Index, SD Standard Deviation, SMD Standardized Mean Difference

Safety

Safety was also comprehensively examined, reviewing both studies that reported adverse events [11, 13, 46, 49, 53, 54, 56, 58, 62–64, 75], and those that explicitly reported no occurrences [12, 29, 41, 59, 70, 76]. BFR training was generally well-tolerated. Among 17 studies (n = 320) reporting safety, minor adverse events were noted, including skin ecchymosis (5 patients, 2 studies) [41, 46], delayed muscle soreness (3 patients, 2 studies) [41, 58], and muscular pain (1 patient, 1 study) [11]. These events were transient, typically resolving spontaneously within days. No serious adverse events, such as thrombosis or cardiovascular complications, were reported.

Safety was managed pre-emptively via strict exclusion criteria. Common contraindications included significant cardiovascular issues (e.g., uncontrolled hypertension [11, 49, 52, 60, 63, 69], comorbidities [11, 52, 62, 63]), a history of thromboembolic disorders (e.g., deep vein thrombosis (DVT), pulmonary embolism) [41, 45, 48, 52, 55, 60, 62, 63, 67, 76], peripheral arterial issues [48, 52, 79], complex peripheral neuropathy [52], and uncontrolled autonomic dysreflexia [63, 76].

Safety monitoring during the interventions was comprehensive (see Supplement Material Table S10). Cardiovascular vital signs were ubiquitously tracked, including blood pressure [42, 50, 51, 53–55, 60, 62, 64] and heart rate [13, 50, 53–55, 60, 62–64]. Subjective feedback was systematically collected via Rating of Perceived Exertion (RPE) or Borg scales [13, 54, 56, 59, 60, 70] and various scales for pain or discomfort (e.g., visual analogue scale, numerical rating scale) [49, 55, 60, 62, 64]. Objective screening involved oximeters for oxygen saturation [13, 42, 51], tissue oxygenation monitoring [62], ultrasound screening for DVT [61, 62], and, in some studies, blood analysis for markers like D-dimer and fibrinogen [50].

Discussion

This study assessed the clinical application of BFR training in neurological disorders, evaluating its efficacy for improving muscle strength and functional outcomes (Fig. 3). Our results demonstrated that BFR training, when added to conventional therapy, significantly enhanced muscle strength, balance (BBS scores), motor function (FMA-LE scores), and activities of daily living (MBI scores). Furthermore, subgroup analysis identified critical factors, such as onset time and intervention duration, that influence effectiveness, while BFR was generally found to have a favorable safety profile.

Fig. 3.

Main findings of the effects of BFR on specific types of neurological disorders from included studies. The inclusion of evidence in the figure required at least one RCT or two or more cases that reported identical dimensional indicators. 1. 1RM/3RM; 2. 10-m Walk Test (10MWT); 3. 12-item Multiple Sclerosis Walking Scale (MSWS-12); 4. 30 Second Sit to Stand Test (30STS/30-sCST); 5. 36-item short-form health survey (SF-36); 6. 5-time Sit-to-Stand test (5STS/FTSTS); 7. 6-Minute Walk Test (6MWT); 8. Activities-specific Balance Confidence Scale (ABC); 9. Berg Balance scale (BBS); 10. DASH (Disabilities of the Arm, Shoulder and Hand); 11. Delayed-Onset Muscle Soreness (DOMS); 12. Expanded Disability Status Scale (EDSS); 13. Fatigue Severity Scale (FSS); 14. Fugl-Meyer Assessment–Balance (FMA-B); 15. Fugl-Meyer Assessment–Lower Extremity (FMA-LE); 16. Fugl-Meyer Assessment–Upper Extremity (FMA-UE); 17. GMFM (Gross Motor Function Measure); 18. Hoehn and Yahr (H&Y); 19. Lower/Upper extremity strength/Handheld dynamometer/MVIC/Torque/Grip strength; 20. Modified Ashworth Scale (MAS); 21. Modified Barthel Index (MBI); 22. Modified Fatigue Impact Scale (MFIS); 23. MS impact scale-29 (MSIS-29); 24. Muscle Mass/Thickness/Volume/Circumference/CSA/ASMI; 25. Pain (NRS/VAS); 26. Parkinson's Fatigue Scale (PFS); 27. Patient-Specific Functional Scale (PSFS); 28. Patient's Global Impression of Change (PGIC); 29. Ratings of Discomfort (RD); 30. Ratings of Perceived Exertion (RPE); 31. Restless Leg Syndrome (RLS) Questionnaire; 32. Spinal Cord Independence Measure (SCIM); 33. Timed 25-foot walk test (T25FW); 34. Timed Up and Go (TUG); 35. Unified Parkinson's Disease Rating Scale (UPDRS); 36. WISCI II (Walking Index for Spinal Cord Injury II) (The figure is created by author YFC)

A key methodological feature of this review was the use of an advanced three-level random-effects model with robust variance estimation (RVE). This approach was critical, as it provided a clear insight into the nature of the high heterogeneity observed in the data [80]. As detailed in the Muscle Strength section, our model confirmed that this variability was driven almost entirely by between-study variance (τ₃²), rather than within-study variance (τ₂²). This finding directly informs the structure of our discussion, and our discussion will first interpret the clinical significance of our core findings on strength and function, and subsequently explore the key moderating factors that explain the high degree of variance observed across studies.

A key advantage of BFR training is its ability to induce fatigue primarily through acute, localized metabolic stress rather than high mechanical tension. While high-load resistance training relies on heavy mechanical loads to recruit high-threshold motor units, BFR achieves similar recruitment patterns by accelerating the accumulation of metabolites within the exercising muscle. This metabolic environment stimulates group III and IV afferents, leading to the compensatory recruitment of fast-twitch motor units even at low intensities [27, 81]. This mechanism serves as a potent physiological stimulus for muscle adaptation, making BFR an effective alternative for individuals with limited mechanical weight-bearing capacity.

This mechanistic distinction is clearly reflected in the recovery profile. The high mechanical load from high-load resistance training frequently causes significant delayed-onset muscle soreness, which often peaks 24 h later [82, 83]. Conversely, BFR training, which minimizes mechanical damage by utilizing lower external intensities, demonstrates a markedly different recovery trajectory. BFR training resulted in significantly lower muscle soreness levels at 30 and 60 min post-training compared to immediately after exercise, indicating a rapid resolution of fatigue and soreness that avoids the characteristic delayed peak associated with high-load resistance training [64]. For clinical populations such as the elderly or obese stroke patients—who often tolerate intense mechanical loads poorly—this profile is a critical advantage, as it reduces the physiological barriers to long-term exercise adherence.

Muscle Strength

Muscle health is considered a critical predictive factor for functional decline and cardiometabolic risk in neurological disorders [84]. Our meta-analysis provides compelling evidence that BFR interventions significantly enhanced muscle strength (p < 0.001) in individuals with stroke [40, 43–45, 48, 50], multiple sclerosis [57], and spinal cord injuries [63]. This benefit was consistent across diverse exercise protocols, including BFR combined with low load resistance exercise [40, 43, 45], aerobic training [57], neuromuscular electrical stimulation [48, 63], repeated transcranial stimulation [44], and transcranial direct current stimulation [50]. This finding is clinically significant, as profound muscle weakness is a pervasive impairment in neurological rehabilitation. The prevalence of weakness in stroke survivors, for example, ranges from 49 to 77% [85]; deficits in muscle size and strength are also observed in both affected and unaffected limbs compared to age-matched controls at the chronic stage [86]. The inadequate development of the musculoskeletal system at both neurological and neuromuscular levels hinders optimal muscle function in individuals with cerebral palsy [16]. Patients with spinal cord injury often experience significant muscle mass loss, with a reduction of up to 40% in muscle cross-sectional area after 6 weeks of injury [17]. Similarly, individuals (n = 354) with multiple sclerosis commonly suffer from leg weakness (90.9%) and fatigue (88.4%) [18], and notable deficits in upper and lower extremities in Parkinson's disease are also well-documented [19]. This widespread presence of weakness underscores the urgent necessity for alternative training modalities that can counteract this impairment, providing a clear opportunity for the application of BFR training.

It is also critical to consider the training protocol design when interpreting these strength results. The literature suggests that low-load BFR training can produce significant strength gains comparable to traditional high-load training, a process largely driven by neural adaptations and metabolic stress [87]. However, achieving these adaptations is often contingent on the protocol being performed to muscular failure. In our review, many of the included trials employed work-matched designs (i.e., prescribing a fixed number of sets and repetitions) rather than a failure-oriented approach. This distinction is vital, as a work-matched, low-load BFR protocol may not impose a sufficient physiological stimulus to maximize strength development compared to a high-load protocol, potentially underestimating the full potential of BFR training.

Functional Outcome Measurement

The strength of the muscles, particularly in the lower extremities, is closely correlated with motor function, gait, and balance in the lower limbs [88]. Our meta-analysis revealed that BFR training not only enhanced muscle strength but also had a positive impact on improvements in balance, FMA-LE scores, and modified Barthel index scores.

This meta-analysis demonstrated a significant improvement in balance following BFR training, a finding that contradicts a previous systematic review [33]. This divergence is likely attributable to the rapid accumulation of evidence in this field; our review captured several newly published studies [41, 43, 46, 47, 55, 57, 77] that were not previously available, providing new support for the effective promotion of balance recovery in patients with neurological diseases. However, it is worth noting that our study only included seven articles related to balance and there was significant heterogeneity in terms of protocols, neurological disorder conditions, and outcome measurements. Sensitivity analysis indicated that Du et al.'s [43] study may have contributed to the observed heterogeneity which surpassed the improvement seen in other six studies [43]. One potential reason for this heterogeneity could be attributed to Du et al.'s [43] focus on training the hip and posterior thigh muscles, which were found to enhance balance but were not considered in other studies [89, 90].

Similar to balance, the analyses for FMA-LE and MBI also demonstrated high levels of between-study heterogeneity (I² > 73%). This suggests that the positive effects, while significant, are highly variable across different studies. Specifically, significant differences in FMA-LE were observed between BFR and control training. However, the evidence supporting these findings remained limited due to the inclusion of only six studies [42–45, 48, 51]. Although the overall results were positive, the high heterogeneity among the included studies regarding intervention duration and onset time made it difficult to determine efficacy. Lastly, meta-analysis also indicated a significant improvement in MBI, but with high heterogeneity mainly attributed to intervention duration.

Determining the other sources of this variability, such as differences in protocols, populations, and onset timing, is crucial for clinical application of BFR. Therefore, these factors were explored in detail in our subsequent subgroup analyses.

Subgroup Analysis

As established in the "Functional Outcome Measurement" section, high between-study heterogeneity was a key finding. Our subsequent subgroup and sensitivity analyses were designed to explore these crucial factors, as determining optimal training parameters and selecting suitable patients for BFR is highly valuable for clinical translation. The subgroup analysis consistently showed that individuals with a shorter time since onset experienced greater improvements in muscle strength and functional recovery. This factor was identified as a primary driver of the heterogeneity observed in outcomes like the FMA-LE, as neuroplasticity, closely associated with disease progression, plays a crucial role in functional recovery of neurological diseases. Typically, patients' spontaneous recovery reaches a plateau 6 months after central nervous system injury [91]. As neurological conditions progress, patients' capacity for neuroplasticity gradually diminishes [12]; thus, they become less responsive to interventions like BFR [11]. For instance, Lamberti et al.'s study [13] demonstrated that BFR had a positive impact on improving gait speed but with a smaller effect size compared to studies [44, 52] employing similar designs but recruiting participants within approximately one month from disease onset [13]. Therefore, early initiation of BFR appears essential for maximizing treatment outcomes.

The findings also indicated that an intervention duration of more than 4 weeks appears essential, particularly for achieving functional improvements. This insight is drawn directly from our subgroup analysis (Table S12), which revealed that for functional outcomes, interventions lasting 4 weeks or less did not reach statistical significance (p = 0.076). In contrast, interventions lasting longer than 4 weeks demonstrated a robust and significant effect (p = 0.001). Interestingly, muscle strength (Table S11) showed a different pattern, with significant improvements observed in both the short-term (≤ 4 weeks, p = 0.035) and long-term (> 4 weeks, p < 0.001) groups. This apparent discrepancy between strength and function aligns perfectly with established principles of training adaptation. A systematic review has shown that strength gains and muscular adaptations are optimized with longer training durations [92]. Short-term programs (often < 6 weeks) typically result in initial strength gains that are primarily driven by neural adaptations. In contrast, longer-duration training allows for more substantial muscular adaptations, such as hypertrophy, to occur alongside these neural changes [93, 94]. This aligns with clinical wisdom, as the typical recommended training duration is often set at 6 weeks to ensure these substantial adaptations are achieved [32]. Therefore, the strength gains we observed within the first 4 weeks may primarily reflect these rapid neural adaptations. Conversely, the more complex functional improvements, which likely depend on substantial muscular adaptations, only became statistically significant after the 4-week threshold. This supports the need for interventions lasting longer than 4 weeks to achieve maximal functional recovery.

An interesting finding from meta-regression and sensitivity analyses was that older age appeared to be a positive moderator, as a non-statistically significant trend suggested that BFR shows promise in mitigating age-related muscle decline [95]. This aligns with a meta-analysis which concluded that BFR training led to superior muscular strength adaptation in older adults, potentially due to its dual effects on muscle and bone metabolism [96]. Indeed, our findings revealed that in studies specifically targeting elderly patient cohorts [40, 43], BFR training demonstrated some of the most significant advantages. Notably, obese patients also exhibited substantial therapeutic benefits [47]. Compared to conventional rehabilitation interventions, BFR may demonstrate superior efficacy in special populations with limited tolerance for high-load exercise training. However, it is worth noting that the participants included in this study had a relatively narrow mean age range (45.3–66.8 years), which necessitates further investigation.

The duration of BFR training sessions typically ranged from 15 to 45 min, with 20 and 30-min durations being the most common. Our study found a trend toward a negative correlation between the duration of single training and functional recovery (Table S13). This may imply that shorter sessions are adequate to provide a sufficient stimulus for adaptation [97] while potentially minimizing excessive systemic fatigue and muscle damage [22]. Although BFR training is reported to induce minimal muscle damage compared to traditional high-load training [22], prolonged BFR training sessions could potentially heighten the likelihood of complications such as numbness or thrombosis [98].

Safety Concerns

Our study demonstrated that BFR training exhibited good overall safety in neurological rehabilitation. Despite concerns regarding hemodynamics and ischemic reperfusion injury, BFR has undergone extensive review and its proper implementation has been confirmed to be well tolerated [56, 59]. This is supported by the absence of exacerbation of original neurological diseases and training-induced injuries [13, 40–42, 46, 49, 53, 62–64]. Studies have also indicated that compared to high-load or low-load resistance training protocols [99], BFR training does not appear to increase muscle damage or injury biomarkers.

Additionally, the level of soreness experienced after BFR training was comparable or even lower than that following high load resistance training [100]. BFR training resulted in significantly lower muscle soreness levels at 30 and 60 min post-training compared to immediately after exercise, while high load resistance training led to significantly higher soreness levels 24 h later than at 60 min post-training.

Limitations

There were several limitations of this study. One limitation was the pooling of all studies and reporting and analyzing all BFR parameters for different neurological diseases. Ideally, a more focused analysis of BFR effectiveness in specific diseases should be pursued. However, this was hindered by the limited number of participants in some articles, lack of randomized controlled trials, and inclusion of only a few case reports. Furthermore, the control interventions in the included studies varied significantly, ranging from passive sham-BFR to active high-load resistance training. We acknowledge that the specific type of control group constitutes a potential source of heterogeneity. Due to the limited number of studies, a stratified subgroup analysis based on control type was not feasible in this review. Consequently, the pooled effect sizes should be interpreted with the understanding that they represent a composite of these varying comparator baselines. Additionally, a key source of the between-study heterogeneity ("Discussion" section) was the heterogeneity in BFR training methods (e.g., cuff pressure, timing, frequencies and duration) with limited studies comparing these factors. Consequently, our meta-analyses were susceptible to type II errors, highlighting the need for high-quality randomized controlled trials to address gaps resulting from moderate to high risks of bias and heterogeneity among studies.

Conclusion

This systematic review and meta-analysis found that utilizing BFR alongside conventional rehabilitation therapy led to greater improvements in muscle strength and functional outcomes in patients requiring neurological rehabilitation. Subgroup and sensitivity analysis revealed that a shorter onset time, longer intervention duration, older age, and shorter duration of single training enhanced the effectiveness of BFR training. Additionally, BFR training was well tolerated by individuals with neurological disorders and had an overall favorable safety profile. However, these findings should be interpreted cautiously and confirmed through more robust studies. Future research should include high-quality randomized controlled trials and comparative effectiveness studies to evaluate the efficacy, safety, and optimal parameters of BFR training.

Abbreviations

BFR

Blood flow restriction

MMT

Manual muscle test

BBS

Berg Balance Scale

5TSTST

Five time sit-to-stand test

TUG

The timed up and go test

6-MWT

6-minute walk test

MBI

Modified Barthel Index

ASMI

Appendicular Skeletal Muscle Mass Index

SMI

Skeletal Muscle Mass Index

WMFT

Wolf motor function test

FMA-UE

Fugl-Meyer Upper Limb assessment

FMA-LE

Fugl-Meyer Lower Limb assessment

SPPB

Short physical performance battery

30sCST

30 Second chair stand test

RM

Repetition maximum

MFIS

The Modified Fatigue Impact Scale

SF-36

The 36-item Short Form Health Survey

T25FW

Timed 25-foot walk test

Author Contributions

HW and JJ accept full responsibility for the work and/or the conduct of the study, had access to the data, and controlled the decision to publish. HW, JJ, YC and LC accessed and verified the data in this study. HW, YC, and LC conducted the database search and data extraction. HW, YC, JZ, LC and YQ conducted the data analyses. All authors interpreted the data, wrote and edited the manuscript. All authors read and approved the final manuscript.

Funding

This study was supported by the National Natural Science Young Foundation of China (No. 82102665), the Shanghai Sailing Program (No. 21YF1404600), the Shaanxi Provincial Health Research Project (2022D071), the Yunnan Provincial Science and Technology Plan Major Project (202203AC100007), the National Natural Science Young Foundation of China (No. 82102666), and the National Natural Science Foundation of China (Grant No. 82272606, 82472602).

Data Availability

The corresponding authors will provide the data supporting the conclusions of this article upon reasonable request.

Declarations

Ethics Approval and Consent to Participate

Not applicable.

Consent for Publication

Not applicable.

Competing interests

The author(s) declare no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.