Does blood flow restriction training enhance clinical outcomes in knee osteoarthritis: A systematic review and meta-analysis

Objective: To systematically review the efficacy of blood flow restriction training (BFRT) on individuals with knee osteoarthritis (OA). Design: Systematic review with meta-analysis. Literature search: Eight electronic databases were searched by one researcher. Study selection criteria: Randomised clinical trials (RCTs) comparing BFRT to regular resistance training (RT) for knee OA. Data synthesis: One reviewer selected the eligible RCTs and exported the data. Two reviewers evaluated study quality using the PEDro scale. We performed meta-analysis where appropriate using a randomeffects model. We rated the quality of evidence using GRADE. Results: Five studies were eligible. The key outcomes analysed were pain, self-reported function, objective physical function, strength and muscle size. Across all comparisons, there was low to moderate quality evidence of no difference between BFRT and traditional RT. Conclusion: The limited available evidence does not suggest that BFRT enhances outcomes for people with knee OA. These findings do not support clinicians using BFRT in people with knee OA. Instead, evidence-based messages regarding exercise and education should remain the mainstay of rehabilitation. Additional studies should clarify whether some people with knee OA who cannot complete an adequate exercise programme due to pain, might still benefit from BFRT to facilitate less painful exercise. © 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

One challenge to long-term adoption of exercise for people with knee OA is that exercise can, especially initially, be painful. Not alone is such pain potentially distressing, but it can reduce compliance among people with knee OA due to concerns that exercise is dangerous (Hendry, Williams, Markland, Wilkinson, & Maddison, 2006). Even though the evidence across various musculoskeletal pain conditions suggests exercise need not be completely painfree to be of benefit (Smith et al., 2017), people with knee OA could benefit from exercising with less pain, especially if it allows them to exercise for longer, or at a greater intensity.
Consequently, BFRT has gained considerable popularity in the last decade as a clinical treatment for painful musculoskeletal conditions. A recent review found that BFRT is a reasonably safe intervention for musculoskeletal disorders (Minniti et al., 2019). Recent studies indicate potentially greater pain reductions in other painful knee conditions within a single session (Korakakis et al., 2018a) and over 8 weeks of training (Giles, Webster, McClelland, & Cook, 2017) compared to a standard strengthening intervention. There have also been studies of BFRT among people with knee OA (Bryk et al., 2016;Segal et al., 2015aSegal et al., , 2015b, however the sample sizes of individual studies have been relatively small, and they have not all reached the same conclusions regarding its effectiveness for people with knee OA. While preparing to publish this review, three other systematic reviews (Cuyul-V asquez et al., 2020;Ferlito, Pecce, Oselame, & De Marchi, 2020;Van Cant, Dawe-Coz, Aoun, & Esculier, 2020) on BFRT for painful knee conditions have been published, one of which focused on knee osteoarthritis (Ferlito et al., 2020). However, some important departures from recommended methodological standards for systematic reviews affect the confidence there can be in these recent reviews, which we highlighted in a recent commentary relating to one of them (Korakakis, O'Sullivan, Whiteley, & Grantham, 2020). Given these ambiguous results, and the large burden associated with knee OA, the aim of this review was to determine the effects of BFRT, relative to regular resistance training, on pain, physical function (self-reported and objective), strength and muscle morphology in individuals with a diagnosis of knee OA or individuals who are identified as being at significant risk of developing knee OA.

Protocol and guidelines
The search strategy and reporting of this systematic review adhered to the PRISMA guidelines (Moher, Liberati, Tetzlaff, & Altman, 2009) and followed recommendations of the Cochrane Handbook for Systematic Reviews (Higgins & Green, 2011). The protocol of the review was prospectively registered in PROSPERO, submitted online in October 2019 (CRD42020154423).

Search strategy
The search was completed in January 2020 using the following electronic databases: Allied and Complementary Medicine Database, Biomedical Reference Collection, CINAHL, MEDLINE, PubMed, PsycARTICLES, PsycINFO, SportsDiscus. Grey literature was searched via OpenGrey, as well as at the following registries: Clinical Trials.Gov and EU clinical trials register. Additionally, reference lists, citation tracking results, and systematic reviews were manually searched to identify studies that were not found through database searching. Search lines were limited to the 'Abstract' and the search terms were limited to synonyms and abbreviations of: 'blood flow restriction training' AND 'knee osteoarthritis' AND 'randomized control trial' OR 'leg degenerative changes'.

Study selection
Upon completion of the database search, each study's title and abstract were collected, imported, and stored in a citation manager. Initially, duplicate studies were removed, followed by the researcher screening titles and abstracts to identify potentially relevant papers. The full-text was then read for any studies that passed initial screening based on viewing the title and abstract, to finalize its eligibility. The search process is shown is based on the PRISMA flow diagram (Fig. 1). (Moher et al., 2009) Study eligibility was determined using the Population, Intervention, Control, and Outcome (PICO) framework (Schardt, Adams, Owens, Keitz, & Fontelo, 2007). Studies were eligible if; (i) participants underwent a variation of BFRT; (ii) participants had been diagnosed with knee OA or were deemed at risk of knee OA; (iii) participants did not have other known co-morbidities; (iv) participants were human; (v) the study performed was a randomized control trial (RCT) or similar randomised comparison (e.g. crossover trial); (vi) the study compared using BFRT to either no intervention, or another intervention not involving BFRT; (vii) the study was written in English. Studies must have involved a BFRT programme of at least 4 weeks duration to be eligible.

Outcomes of interest
Pain intensity, self-reported functional ability or quality of life, objective physical function, lower limb strength and muscle volume were the key outcomes of interest.

Data extraction
Data was extracted and cross-checked from each eligible study by one researcher (BG). The following data was extracted from each study: (1) participant characteristics: age, sex, body mass index (BMI), and Kellgren-Lawrence OA grade; (2) intervention characteristics: blood flow restriction (BFR) cuff type and pressures, intervention type and frequency and duration; and (3) data regarding relevant outcome measures. Data were extracted and exported to Microsoft Excel, before being assessed for similarities in participant characteristics, interventions, and outcome measures.

Quality assessment
Eligible studies identified were critiqued using the Physiotherapy Evidence Database (PEDro) scale by two reviewers, which determines any potential risks for bias within a study and has been B. Grantham, V. Korakakis and K. O'Sullivan Physical Therapy in Sport 49 (2021) 37e49 established as a reliable tool for assessing RCT's (Maher, Sherrington, Herbert, Moseley, & Elkins, 2003). The scores were confirmed by cross-checking with the scores awarded on https:// pedro.org.au/. The PEDro scale consists of 10 questions which identify potential weaknesses within each study (Maher et al., 2003). Questions one and two target participant group allocation and randomization, while questions five, six, and seven explore the types of blinding performed within RCT's (Maher et al., 2003). The remaining questions (four, eight, nine, and ten) assess participant characteristics and methods of reporting results (Maher et al., 2003) and were a crucial focal point when comparing the eligible studies.

Data analysis, synthesis and summary of findings
Outcome data were transformed to ordinary 0e10 and 0e100point scales for pain and function where applicable, respectively. As measures of treatment effect, we calculated and presented standardised mean differences (SMDs) and 95% confidence intervals (95%CI). Where possible outcome data were pooled, and heterogeneity was not judged only by the value of I 2 statistic, as thresholds for the interpretation can be misleading (Schroll, Moustgaard, & Gotzsche, 2011). Since clinical and methodological diversity always occur in quantitative synthesis, statistical heterogeneity is inevitable (Higgins, Thompson, Deeks, & Altman, 2003). Statistical heterogeneity was assessed as follows (Ioannidis & Trikalinos, 2007): (1) overlap (poor or adequate) of CIs presented in forest plots; (2) magnitude and direction of effects; (3) sample sizes and number of studies included (as small number of participants and/or studies included in analysis results in low power of heterogeneity test); and (4) strength of evidence for heterogeneity (p value from c 2 test or CI for I 2 ) (Higgins & Green, 2011;Schroll et al., 2011). Based on the characteristics of the included studies we assumed that clinical and methodological heterogeneity was likely to exist and to impact the outcomes; hence we used a random-effects model to pool outcomes. All analyses were conducted using Comprehensive Meta-Analysis software, V3 (Biostat, Englewood, New Jersey, USA).
Two reviewers assessed the quality of the current evidence using Grading of Recommendations, Assessment, Development and Evaluation (GRADE) methodology (Guyatt et al., 2011). We prepared tables summarizing the findings and assessed the quality of the evidence as "high", "moderate", "low" or "very low" depending on the presence and extent of: risk of bias (mean PEDro Score was <5 out of 10) (Mendonca et al., 2020); inconsistency of effect (based on above criteria for heterogeneity); indirectness (downgrade if clinically heterogeneous); and imprecision (downgrade if upper or lower confidence interval spanned an effect size of 0.5 in either direction). Assessment of publication bias was not possible due to the small number of included trials (Jansen, Viechtbauer, Lenssen, Hendriks, & de Bie, 2011). We a-priori graded an outcome with only one trial as low quality, and if it also had high risk of bias the evidence was graded as very low quality (Atkins et al., 2004).
We undertook subgroup analyses to compare the effect of exercise with or without BFRT on knee OA in trials that used (i) high or low intensity exercise as a comparator and (ii) between studies that included only male or female participants.

Sensitivity analyses
We aimed to repeat the meta-analyses by excluding studies with poor quality, studies appearing as outliers, and by using weighted values in studies reporting subgroup comparisons and used the same control comparator to assess their influence on the pooled effect. Also, given that the I 2 statistic provides the proportion of the observed variance that can be attributed to the variance in true effects rather than to sampling error, we also calculated the prediction interval to evaluate the true effect size range in the metaanalyses (Borenstein, Higgins, Hedges, & Rothstein, 2017).

Study selection and participant characteristics
The search strategy identified a total of 3238 studies, including 601 duplicates. After duplicate removal, 87 studies passed initial screening based on viewing the title and abstract, and these fulltexts were assessed for inclusion in this review. The flow of trial identification, screening, and eligibility assessment process is presented in Fig. 1.
Notable reasons for exclusion from the review were the absence of a true BFRT intervention, a non-RCT design, or the study did not include individuals with (or who were "at risk" of) knee OA.
Study characteristics such as, sample size, age, gender, length of follow-up, interventions, outcome measures, main results, and study quality are presented in Table 1. All included studies were published in English and were carried out in 2 countries, USA (3 trials) and Brazil (2 trials).
The eligible studies included 199 participants: 147 females and 52 males, with a mean age of 60.3 (range 54.6e69.1) years. The median number of participants randomised per trial was 42 (IQR 34.5e46.5) and the sample size ranged from 34 to 48. Three studies included participants with previously diagnosed knee OA (n ¼ 117) (Bryk et al., 2016;Ferraz et al., 2018;Harper et al., 2019), and two studies' participants were defined as "at risk" of knee OA (n ¼ 82) (Segal et al., 2015a(Segal et al., , 2015b. In addition, the baseline characteristics between control and BFRT groups did not differ in all five studies, although prior to adjusting for BMI, one study (Nielsen et al., 2012) reported the BFRT group had a significantly lower BMI than the control group. All studies reported only short-term effects of BFRT (4e12 weeks).
Diagnostic criteria for knee OA used by two studies (Bryk et al., 2016;Ferraz et al., 2018) were the clinical and radiographic criteria as established by the American College of Rheumatology (Arthritis Foundation, 2019); one study (Harper et al., 2019) defined the presence of knee OA by (1) radiographic evidence of osteophytes, (2) pain classification > grade 0 on Graded Chronic Pain Scale, and (3) bilateral standing anterioreposterior radiograph demonstrating Kellgren and Lawrence grade 2 of the target knee; while two studies (Segal et al., 2015a(Segal et al., , 2015b included participants that either had radiographic knee OA without symptoms or had at least 1 of the following risk factors for symptomatic knee OA; (1) knee injury resulting in inability to walk without assistance for at least 2 days; (2) knee surgery (other than bilateral knee arthroplasty); (3) knee pain, aching, or stiffness on most of the prior 30 days; (4) or were overweight or obese (BMI >25 kg/m 2 ).

Intervention characteristics
The duration of interventions ranged from 4 to 12 weeks, with two or three exercise sessions per week. The control groups performed relatively equivalent exercises to the BFRT group, and the training load was based on 1RM. The control group training intensities ranged from low intensity (LI) to high intensity (HI) (30e80% of 1-repetition-maximum (1-RM)), while BFRT group training was always LI (20e30% of 1-RM). Of the five studies, only one (Ferraz et al., 2018) used a 3-arm RCT design with a BFRT intervention group and two control groups: a LI resistance training group and HI resistance group. Exercises performed by BFRT and control groups throughout all five studies primarily focused on knee extensor muscles using a leg-press exercise or knee extension with conventional resistance training machines. However, two studies (Bryk et al., 2016;Harper et al., 2019) included lower-body general strength training along with flexibility exercises and balance training.
All studies provided the name of the exercise intervention, along with information on repetitions and sets. Two studies (Ferraz et al., 2018;Harper et al., 2019) provided a detailed description of loading progression, two studies (Segal et al., 2015a(Segal et al., , 2015b did not adjust the training load during training period, while one study (Bryk et al., 2016) did not provide sufficient details. Three studies (Bryk et al., 2016;Segal et al., 2015aSegal et al., , 2015b provided rest time between sets, two studies (Harper et al., 2019;Segal et al., 2015b) described time under tension, all studies involved or supervised training and reported who provided the intervention, three studies (Ferraz et al., 2018;Harper et al., 2019;Segal et al., 2015a) gave information about fidelity or adherence, and no study described motivation strategies.
Substantial variability in BFRT cuffs and the precise cuff pressure used during exercise was present among studies and are outlined in Supplementary file 1.

Adverse events
Two studies (Ferraz et al., 2018;Harper et al., 2019) reported adverse events during training. The majority of these events were related to exercise-induced knee pain, with the BFRT group reporting less than the control group, though the only serious adverse event reported was within the BFRT group (Harper et al., 2019).

Outcome measures
Substantial variation was evident in the outcome measures used in the included studies. A total of nineteen outcome measures were used.
Objective physical function was assessed in four studies (Ferraz et al., 2018;Harper et al., 2019;Segal et al., 2015aSegal et al., , 2015b) using a range of tests (i.e. Timed-Up and Go e TUG, stair climb muscle power, time-stands test, walking speed).
3.6. Effectiveness of BFRT compared to resistance training alone: quantitative meta-analyses 3.6.1. Included and excluded studies Four studies (Bryk et al., 2016;Ferraz et al., 2018;Segal et al., 2015aSegal et al., , 2015b were included when pooling data, as one study (Harper et al., 2019) was not powered to detect statistically significant differences in outcomes (only estimated mean differences with 95% CIs were reported).

Effect of intervention -knee pain
There was a moderate level of evidence suggesting no difference in pain reduction between LI BFRT and resistance training alone at short-term follow-up (Table 3, Fig. 2A). The evidence was downgraded due to clinical heterogeneity.

Effect of intervention e self ereported functional disability
A low level of evidence suggested no difference in functional disability between LI BFRT and resistance training alone at shortterm follow-up (Table 3, Fig. 2B). The evidence was downgraded due to clinical heterogeneity and imprecision.

Effect of intervention e objective physical function (mobility and balance)
A moderate level of evidence suggested no difference in physical mobility and balance (TUG test in seconds) between LI BFRT and resistance training alone at short-term follow-up (Table 3, Fig. 2C). The evidence was downgraded due to imprecision.
Similarly, individual study results indicate that neither the timeto stand nor stair climb power significantly improved after the exercise interventions (twelve and four weeks, respectively), for BFRT or control treatment groups (Ferraz et al., 2018;Segal et al., 2015b).  Note: Data from one study (Ferraz et al., 2018) was calculated and extracted from figure.; Abbreviations: SMD, standardized mean difference, CI, confidence intervals; BFRT, blood flow restriction training; TUG, timed up and go test. a Clinical heterogeneity. b Imprecision (lower confidence interval spanned an effect size of 0.5).
B. Grantham, V. Korakakis and K. O'Sullivan Physical Therapy in Sport 49 (2021) 37e49 3.10. Effectiveness of BFRT compared to resistance training alone: qualitative analyses Substantial heterogeneity in reporting did not allow quantitative synthesis of the results for strength and muscle size.

Effect of intervention e muscle strength
Three of the five studies reported no significant difference in strength gains between BFRT and conventional RT (Bryk et al., 2016;Ferraz et al., 2018;Segal et al., 2015a). The other two studies reported very small differences, in opposite directions, with one study (Segal et al., 2015b) indicating greater strength gains with BFRT while the other study (Harper et al., 2019) suggested inferior strength gains with BFRT. These small contradictions may well simply reflect differences in the intensity of exercise performed in the control arms.

Effect of intervention e muscle size
Neither of the two studies reported significant between-group differences in muscle size. Only the study with the 12-week follow-up reported significant within-group increases in quadriceps cross sectional area in the HI-RT (þ8%, ES ¼ 0.54, P < 0.0001) and the BFRT group (þ7%, ES ¼ 0.39, P < 0.0001), but not in the LI-RT group (þ2%, ES ¼ 0.12, P ¼ 0.52).

Effect of intensity of resistance training
In subgroup analysis we evaluated the effect of low intensity BFRT compared to both high or low intensity resistance training in pain, self-reported functional disability, and objective physical function (mobility and balance) (Table 4, Fig. 3). Three studies used as a comparator a LI-RT group (Ferraz et al., 2018;Segal et al., 2015a, Fig. 2. Forest plots for the effectiveness of BFR with low intensity resistance training compared to low or high intensity resistance training alone in patients with knee osteoarthritis. Data is depicted according outcome. A) Pain reduction at short-term follow-up, B) Functional disability at short-term follow-up, and C) Physical mobility and balance at short-term follow-up. Abbreviations: SMD, standardized mean difference; SE, standard error; CI, confidence intervals; BFRT, blood flow restriction training; RM, repetition maximum; TUG, timed up and go test; M, male, F, female. 2015b), while two studies used as a comparator a HI-RT group (Bryk et al., 2016;Ferraz et al., 2018).
Finally, a moderate level of evidence suggests no difference in physical mobility and balance (TUG test in seconds) between LI BFRT and HI resistance training alone (SMD ¼ 0.056, 95%CI: À0.427, 0.539) at short-term follow-up (Table 4, Fig. 3D). Low level of evidence suggests no difference in physical mobility and balance (TUG test in seconds) between LI BFRT and LI resistance training alone (SMD ¼ À0.585, 95%CI: À1.293, 0.123).

Effect of interventions according to gender
We performed a subgroup analysis by gender to evaluate the effect of interventions in pain. One study included only male participants (Segal et al., 2015a), while three studies recruited female participants (Bryk et al., 2016;Ferraz et al., 2018;Segal et al., 2015b).
A moderate level of evidence (downgraded due to clinical heterogeneity) suggests no difference in pain reduction between BFRT and resistance training alone in women with knee OA (SMD ¼ À0.076, 95%CI: À0.410, 0.258) at 4 weeks (Fig. 3E). A low level of evidence suggests no difference in pain reduction between BFRT and LI-RT alone in men (SMD ¼ 0.363, 95%CI: À0.256, 0.981).

Sensitivity analyses
We considered no study as being poor quality, or appearing in the analyses as an outlier. Using weighted values in studies reporting subgroup comparisons by using the same control group, did not significantly affect the direction or the size of the pooled effect estimate, while subgroup analyses resolved between-study variability in the CIs of effects.
Finally, the prediction intervals calculated (assuming that the effects were normally distributed) revealed that the true effect size for any single population with knee OA in all outcome measures will usually fall within a greater range than that reported in the present systematic review (Supplementary file 2).

Main findings
This systematic review found no difference, albeit based on low to moderate quality evidence, between BFRT and traditional RT for any of the clinical outcomes examined in people with knee osteoarthritis. As such, these findings do not support the current use of BFRT in the rehabilitation of knee OA.
There is no question that exercise can reduce pain and enhance function in people with knee osteoarthritis, both in the short-term and medium-term (Fransen et al., 2015;Lange, Vanwanseele, & Fiatarone Singh, 2008). A more pertinent question for clinicians, and patients, is whether other therapies or adjuncts might further Note: Data from one study (Ferraz et al., 2018) was calculated and extracted from figure.; Abbreviations: SMD, standardized mean difference, CI, confidence intervals; BFRT, blood flow restriction training; HI-RT; high intensity resistance training, LI-RT; low intensity resistance training, TUG, timed up and go test. a Imprecision (upper or lower confidence interval spanned an effect size of 0.5). Fig. 3. Subgroup analyses forest plots for the effectiveness of BFR with low intensity resistance training compared to resistance training alone in patients with knee osteoarthritis. Data is depicted according outcome. A) Pain reduction BFRT compared to LI-RT at short-term follow-up, B) Pain reduction BFRT compared to HI-RT at short-term follow-up, C) Functional disability BFRT compared to HI-RT at short-term follow-up, D) Physical mobility and balance BFRT compared to HI-RT at short-term follow-up, and E) Pain reduction BFRT compared to LI-RT only in women at short-term follow-up. Abbreviations: SMD, standardized mean difference; SE, standard error; CI, confidence intervals; BFRT, blood flow restriction training; RM, repetition maximum; TUG, timed up and go test; HI-RT, high intensity resistance training, LI-RT, low intensity resistance training, F, female.
enhance the known benefits of exercise. Our findings suggest outcomes for people with knee osteoarthritis are not further enhanced by BFRT. This ineffectiveness is consistent with the evidence for many other adjuncts to exercise in painful musculoskeletal conditions (including insoles, taping, manual therapy, electrotherapy) (Collins et al., 2018;Karabulut, Mccarron, Abe, Sato, & Bemben, 2011;Logan et al., 2017;Zhang, Wang, & Zhang, 2018) where greatest benefits are apparent when they are used without accompanying exercise. Perhaps reflecting the current interest in BFRT, on completion of this review but before publication, three other systematic reviews (Cuyul-V asquez et al., 2020;Ferlito et al., 2020;Van Cant et al., 2020) on BFRT for painful knee conditions were published, one of which focussed on knee osteoarthritis (Ferlito et al., 2020). There was considerable overlap between the studies included across these three reviews and our current review, though the precise eligibility criteria varied. We recently published a critique of one of the reviews (Korakakis et al., 2020) which extracted data inaccurately, and pooled a wide range of pain intensity data in a manner which is open to question. Looking across all these reviews, it appears clear that; (i) overall, BFRT does not enhance the key clinical outcomes of pain and disability relative to regular HI-RT; (ii) there may be value in exploring the role of BFRT in situations where HI-RT is not possible e.g. due to excessive pain, though the data for this is not yet strongly established. These findings are also relatively consistent with the most promising results reported for BFRT in other painful musculoskeletal conditions. For example, Giles et al. (2017) (Giles et al., 2017) which was included within two of the systematic reviews (Cuyul-V asquez et al., 2020;Van Cant et al., 2020) on painful knee conditions, but not in the current review reported that 'pain with daily activities' improved significantly more for people with patellofemoral pain using BFRT than those undergoing only traditional RT. However, this additional benefit was only evident immediately post-intervention at eight weeks, and was no longer evident at six months. Secondly, and arguably more importantly, pain with daily activities was not a primary outcome in that RCT, with neither the two primary outcomes (worst pain, and pain-related function) nor any other secondary outcomes demonstrating significant benefits for the BFRT group. Our findings are also consistent with Ladlow et al. (2018) (Ladlow et al., 2018) who reported no significant benefit of BFRT over conventional RT among people with lower limb musculoskeletal pain. We are aware of no other RCTs demonstrating a sustained (e.g. at least 4 weeks) clinical benefit on pain or disability for BFRT, over and above the benefit of exercise, among people with painful musculoskeletal conditions.

Clinical implications
The most promising aspect of BFRT for painful conditions is that it might allow more intense exercise with less pain, and/or reduce within-session increases in pain associated with exercise. While there is preliminary evidence to support this (Giles et al., 2017;Korakakis et al., 2018bKorakakis et al., , 2018c, any such differences in pain are not always statistically significant. While within-session pain reports were not a key outcome for our review, it is worth noting that less people discontinued their exercise due to pain in the BFRT group in one of the five studies (Ferraz et al., 2018). Furthermore, greater within-session soreness associated with the exercises was reported in another two studies (Bryk et al., 2016;Harper et al., 2019). We do not suggest such pain experiences are irrelevant, as ideally patients would not experience any more pain that necessary during rehabilitation. However, any potential benefits related to this withinsession pain relief did not result in better clinical outcomes.
A critical consideration here is whether it is necessary and beneficial, as opposed to desirable, for patients to have less pain during exercise. A recent systematic review (Smith et al., 2017) found no evidence that painfree exercise results in better outcomes long-term than exercise involving pain among people with painful musculoskeletal conditions. This is also consistent with contemporary evidence regarding pain mechanisms, in that pain being experienced does not always mean harm is being done ('hurt is not harm') (Butler & Moseley, 2003). In that regard, it would seem sensible to encourage exercise and simply monitor the painresponse to exercise over time (Domenech, Sanchis-Alfonso, L opez, & Espejo, 2013;Silbernagel, Thome e, Eriksson, & Karlsson, 2007) to determine if the exercise needs to be modified to ensure patient comfort, and maximise compliance. Rather than being part of a regular protocol, it is perhaps only in the event that pain is deemed excessive by the patient during exercises, that there might be a role for adjunct therapies to ease within-session pain. BFRT is one such adjunct, and the choice of clinicians and patients might depend on the importance placed on being able to achieve a high intensity stimulus without as much effort being required by the patient, versus the complexity and cost of adding BFRT to a simple exercise programme. There was no evidence of significant harm associated with BFRT in the eligible studies. While reports of serious adverse events (e.g. embolism) are rare, they do occur (Minniti et al., 2019;Nakajima et al., 2006), and are possibly more likely in populations presenting with OA due to their age. In this regard, there are perhaps parallels with cervical manipulation for neck pain, where the overall risk is low, but potentially very serious if it occurs. For these rare examples of when conservative rehabilitation can cause serious danger, the importance of proving efficacy becomes even greater. Further research into how such rare adverse events can be minimised would be valuable.
We appreciate a real challenge for clinicians is meeting patient expectations, and indeed funding pressures, in an evidence-based manner. It can be easier to offer a novel therapy, and charge a fee commensurate to such novelty, than to persist with the evidencebased message that chronic health conditions such as knee OA require long-term behavioural change such as investing in changing activity patterns, exercising, altering diet and managing mental health. In this regard, we found it interesting that three of the five included studies actually portrayed their findings in a positive manner in their conclusions, typically with statements such as BFRT being 'similarly effective' to RT. If a clinician is to add another therapy to the management plan for a person with knee OA, considering the potentially extra costs, risks and burdens, it should surely improve existing outcomes, not merely match them.

Strengths and limitations
This review was registered prospectively, analysed a wide range of outcomes and has been reported in accordance with recommendations. We acknowledge the quality of evidence is only low to moderate, including only five studies with relatively small samples. Importantly, only three of the five studies included people with established knee osteoarthritis. None of the trials were adequately blinded, though this is common for exercise trials. There was considerable variation between the BFRT protocols used in the studies, in terms of the frequency (2e3 sessions per week) and duration of the training programme (four to 12 weeks), the precise training pressure used, whether the pressure was expressed relative to the pressure needed for arterial occlusion, the duration of a single BFRT session and BFRT cuffs used (See Supplementary file 1). While such variation leaves open the possibility that more consistent application of a specific BFRT protocol could be more effective, none of the various protocols showed clear benefits in the eligible studies. In three studies (Bryk et al., 2016;Ferraz et al., 2018;Harper et al., 2019) the control group exercised at a higher intensity than the BFRT group. While this disadvantages the BFRT group to some extent, this is the very premise upon which BFRT for painful conditions is based i.e. that it can mimic higher-intensity training with less pain and/or greater adherence (Abe et al., 2005a;Horiuchi & Okita, 2012;Takarada et al., 2000). It is possible RCTs in languages other than English were missed.

Conclusion
Across all comparisons, this systematic review found low to moderate quality evidence of no difference between BFRT and traditional RT for any clinical outcome. While the conclusions of this review need to be tempered by the small number of studies, and the varying outcome measures and protocols employed, the findings do not support the use of BFRT as the default choice in rehabilitation programmes for knee OA. In the event that some people with knee OA cannot complete an adequate exercise programme due to pain, BFRT may be an option to facilitate less painful exercise.

Key points
BFRT has been advocated as a potential mechanism to improve clinical outcomes in painful musculoskeletal conditions by allowing more challenging exercise to be performed with less pain. In this systematic review, there was low to moderate quality evidence of no difference between BFRT and traditional RT for any clinical outcome. BFRT does not enhance clinical outcomes for people with knee OA, though there may be occasions when BFRT could be considered such as if people with knee OA cannot complete an adequate exercise programme due to excessive pain.

Ethical statement
Systematic review of previously published data-no ethical approval sought.

Grant support
None.

Financial disclosure and
Nothing to declare.

Institutional review board approval
Not required.

Declaration of competing interest
No funding was obtained for this review.