Summarizing the effects of different exercise types in chronic low back pain – a systematic review of systematic reviews

Background In chronic LBP (CLBP), guideline-endorsed treatment is to stay active, return to normal activity, and to exercise. Several reviews on various exercise types used in CLBP have been published. We aimed to identify systematic reviews of common exercise types used in CLBP, to appraise their quality, and to summarize and compare their effect on pain and disability. Methods We searched the databases OVID MEDLINE, EMBASE, COCHRANE LIBRARY, and WEB OF SCIENCE (Core collection) for systematic reviews and meta-analyses on adults between 18 and 70 years of age suffering from chronic or recurrent LBP for a period of at least 12 weeks, which investigated the effects of exercises on pain and disability. All searches were conducted without language restriction. The search was performed up until 2022–01-26. The included reviews were grouped into nine exercise types: aerobic training, aquatic exercises, motor control exercises (MCE), resistance training, Pilates, sling exercises, traditional Chinese exercises (TCE), walking, and yoga. The study quality was assessed with AMSTAR-2. For each exercise type, a narrative analysis was performed, and the level of evidence for the effects of exercise was assessed through GRADE. Results Our database search resulted in 3,475 systematic reviews. Out of the 253 full texts that were screened, we included 45 systematic reviews and meta-analyses. The quality of the included reviews ranged from high to critically low. Due to large heterogeneity, no meta-analyses were performed. We found low-to-moderate evidence of mainly short-term and small beneficial effects on pain and disability for MCE, Pilates, resistance training, TCE, and yoga compared to no or minimal intervention. Conclusions Our findings show that the effect of various exercise types used in CLBP on pain and disability varies with no major difference between exercise types. Many of the included systematic reviews were of low-to-moderate quality and based on randomized controlled trials with high risk of bias. The conflicting results seen, undermine the certainty of the results leading to very-low-to-moderate quality of evidence for our results. Future systematic reviews should be of higher quality to minimize waste of resources. Trial registration PROSPERO: Reg no 190409 Registration date 01AUG 2020. Supplementary Information The online version contains supplementary material available at 10.1186/s12891-022-05722-x.

One reviewer per pair extracted data from the included reviews and the other reviewer from the same pair checked the extraction for accuracy. We extracted the data into a data extraction form, adapted from a Cochrane form (27). We extracted data primarily at the systematic review level, but if necessary, the original RCTs were checked for further details. The data were synthesized by presenting PICO. The results of each included systematic review were separated on the primary outcomes, pain and disability, if possible. We did not perform a meta-analysis, since clinical homogeneity was not present due to the large variation in exercise dosages, combinations of interventions, differences between the studies in controls groups as well as outcome measures, and follow-up times.

Assessment of certainty of evidence
We used the GRADE approach (47) to evaluate certainty of the level of evidence for each exercise type and primary outcomes. In this systematic review, we used the conclusions by the authors as the main source, but we also checked if the results were statistically signi cant compared to a control intervention. When possible, we also used the established minimal important difference (MID) as a speci ed threshold in our evaluation of the level of evidence. In short, the rst step of GRADE is to choose the start point of the level of evidence, and since we included (mostly) RCTs we decided to set this at the highest level. Thereafter, we lowered this level of evidence by appraising the potential limitations due to study limitations (high risk of bias/AMSTAR points), inconsistency (in results), imprecision (large con dence intervals, heterogeneity), indirectness (poor measurement quality), and publication bias. The level was increased if large effects or a "dose-response" were seen based on the reports of the systematic reviews. We did not include residual confounding factors for upgrading, due to the inclusion criteria to include mainly RCTs. In this way, we express our ndings together with the con dence on the results, using four levels of evidence: "high" (++++), "moderate" (+++), "low" (++) or "very low (+) (31).

Results
The search results are summarized in Fig. 1. The literature search returned a total of 2,345 studies. Following removal based on duplicates, a review of title and abstracts (n = 1654) was performed, and 246 full texts were screened. After checking against our inclusion and exclusion criteria, we included in total 41 publications in the nal review. A list of excluded studies and reasons is included in APPENDIX 3.
Except for resistance training (CCA 4%), in all exercise types with more than one review included, there was a high or very high overlap of the included studies (walking: CCA 38%).

Summary results
The narrative analyses of included SRs/MAs showed large effects when comparing the exercise interventions with minimal or no intervention and are summarized in table 2. For most exercise types, there were no differences when different exercise types were compared with each other. Mostly small or nonsigni cant effects on pain and disability were found in favour of the various exercise types compared with other control interventions, such as usual care. We found low to moderate evidence that any exercise type is effective for reducing pain and disability compared to no or minimal intervention, but that no exercise type is more effective than another. a) Aerobic exercise Aerobic exercises aim to improve the e ciency and capacity of the cardiorespiratory system (70). Our search resulted in one meta-analysis on the effects of aerobic exercise which covered a literature search up to March 2016 and included six studies with 333 subjects (36). The methodological quality varied across the studies. The review reached 13 points on AMSTAR-2 and was rated as having high quality (Table 1). Aerobic exercise was compared to resistance training, or combined aerobic and resistance training versus exercise advice, maintained normal activity, or waiting list (APPENDIX 4).
The results showed that aerobic exercise reduced pain, although neither aerobic nor resistance training proved to be superior to the other (Table 3a). No signi cant differences were reported for disability.
The GRADE analysis showed moderate quality evidence that aerobic exercise is as effective for the reduction of pain and disability compared to resistance training (Table 4). We downgraded due to possible publication bias, since only one review was identi ed. b) Aquatic exercise Aquatic exercises are any exercises performed in water, such as running, active range of motion, or strengthening (50). The literature search identi ed only one SR about the effectiveness of aquatic exercises, compared to land-based or no exercises (50). The review included eight RCTs with a total of 311 participants (Table 3b). The review was published in 2018, conducted its search in November 2016, and included a meta-analysis. The included RCTs were of moderate to high quality. The review reached 9 out of 16 points on AMSTAR-2, indicating moderate quality ( Table 1). The reasoning of the study design was not reported, no list of excluded studies was provided, no reporting of funding, and RoB was not considered in the MA.
The MA found a statistically signi cant, but clinically questionable, reduction in pain and disability in patients treated with aquatic therapy compared to patients treated with land-based therapy (Table 3b). No information about the time point of outcome reporting was provided (APPENDIX 4).
The GRADE analysis showed that there is low quality evidence that aerobic exercise is superior in reduction of pain and disability compared to land-based exercise ( Table 4). The evidence was downgraded due to study limitations and possible publication bias, since only one review was identi ed.

c) Motor control exercises
Motor control exercises aim to restore the neuromuscular control of the muscles stabilizing the spine, and are graded from low load exercises into activation during functional exercises and activities (71). We included 11 reviews (18,35,40,42,45,46,48,(55)(56)(57)65) on motor control exercises (MCE), including one review on movement control exercises (45). All but one (57) had conducted a meta-analysis. The publication year ranged from 2008 to 2020 and the last updated search was October 2018 (48). The reviews included between four to 34 low to high quality RCT's and included between 209 to 2514 participants. There was a high overlap of the included studies (CCA 14%). AMSTAR-2 ranged from 4 to 16 points out of 16 (Table 1). Only two of the reviews were rated with an overall high quality (18,35). Several of the reviews reported "nearly yes" on the item "presenting a protocol", and most of the reviews reported a "No" on items "presenting a list of the excluded studies" and "presenting the source of funding and con icts of interest".
The reviews reported outcomes of pain and disability in the short, intermediate and long-term (Table 3c). Control interventions were general exercises (GE), spinal manual therapy (MT), multimodal treatment (MMT), or information/minimal intervention/usual care (MI) (APPENDIX 4). The narrative synthesis on pain in the included reviews showed a small but non-signi cant effect for motor control exercises over general exercises in some of the reviews mainly in the short and intermediate term (18,35,40,42,55,56,65). Compared to manual therapy, none of the nine reviews presented any results on differences to motor control exercises for pain (18,35,40,42,45,46,55,56,72). Four reviews, however, reported signi cant and clinically relevant results showing that motor control exercises were more effective in short, intermediate and long-term compared to minimal intervention (18,40,46,55).
The narrative synthesis on disability in the included reviews showed a small but nonsigni cant effect for motor control exercises over general exercises (18,35,40,42,46,55,65), while Niederer et al. (2020) presented results of no difference at any time points (48). Motor control exercises showed small and nonclinically relevant results compared to manual therapy in two reviews (40, 46). Compared to minimal intervention motor control exercises showed signi cant and clinically signi cant differences in the short (18,40,55), intermediate (18,40,46,55), and long-term (18,40).
The GRADE analyses showed that there is a moderate level of evidence on the effect of motor control exercises on pain compared to minimal intervention, and a low level of evidence that there is an effect on disability (Table 4). Downgrading was based on the inconsistency of the results, and for disability also on imprecision due to signi cant heterogeneity.

d) Resistance training
Resistance training includes exercises to improve the strength, power, endurance and size of skeletal muscles (70). The interventions included resistance training, back muscle training and medical training therapy. We included three SR's on the effect of resistance training (66,67,69), but none conducted a metaanalysis. Publication year ranged from 2001 to 2012 and the last updated search was performed in April 2010. In one review, only two RCTs were included (67), while the other two reviews included 12 and seven RCTs, respectively, with a small overlap (CCA 4%) (66,69). Two reviews reported moderate to high quality of the included RCTs (66,67), while one review did not report on study quality (69). AMSTAR-2 scores were very low and ranged from 1.5 to 5, indicating critically low, and low-quality reviews (Table 1). Resistance training was compared with passive treatments, tness training, no treatment placebo, or cognitive interventions (APPENDIX 4).
All included reviews reported on decreased pain scores in the included studies compared to passive or no intervention (66,67), but it was unclear if the effect was clinically relevant or what time-period was used. One review found no difference compared to a cognitive behavioural intervention, and the effect disappeared at the long-term follow-up (67), while another review reported no difference in pain scores when compared to tness training (66) (Table 3d). Resistance training was found effective for the reduction of disability in all reviews compared to passive or no intervention in one review (66), but it was unclear what comparison groups were used in the other two (67,69) (Table 2d).
The GRADE analyses showed that there was a very low level of evidence that resistance training has positive effects on pain and disability, but not compared to tness training and cognitive behavioural intervention ( Table 4). The level of evidence was downgraded due to low study quality, inconsistency, imprecision, and an increased risk of publication bias.

e) Pilates
Pilates exercises follow the traditional Pilates principles such as centring, concentration, control, precision, ow and breathing (52).The literature search resulted in nine systematic reviews (19,32,39,44,47,52,61,62,64), of which ve had performed an MA on the effect of Pilates (19,32,44,47,62). The reviews included between four to 14 RCTs, published between 2005 and 2016, and included between 134 (62) to 708 participants (39). There was a very high overlap of the included studies (CCA 32%). Publication years ranged from 2011 (44, 64) to 2018 (39); the last updated search was April 2016 (39). The studies included in the reviews were graded with different types of quality scores and were rated as being of moderate quality, with some exceptions to both low and high quality. The study quality ranged from low, 7 points (64) to 16 points (19) out of 16 points of AMSTER-2 ( Table 1). Most of the reviews did not present the source of funding or con ict of interest (Item 10). All studies with MAs signi cant large heterogeneity. The intervention dosage varied greatly between the studies, and due to poor reporting, it was impossible to summarize a typical exercise duration, frequency, or intensity. The control interventions also varied greatly between the different studies, and contained treatments such as other exercise types, McKenzie, massage, back school programs, or information/minimal intervention/usual care (APPENDIX 4).
The narrative synthesis on the outcome pain showed signi cant (but mostly small) effects for Pilates over no or minimal intervention in eight reviews (19,32,39,44,47,52,61,64). In all included reviews there were no differences compared to other types of exercises, except for one review that found a superiority for Pilates exercises compared to physical activity (52). Similar results were found in the narrative synthesis on disability. Six of the included reviews reported on non-signi cant effects for Pilates over minimal intervention (19,39,47,52,61,64), and the majority of the reviews pointed out that Pilates exercises were as effective compared to other types of exercises, mainly with short-time effects (Table 3e). Most outcomes were of short or intermediate term and most often the effect was lower than the recommended minimal clinical important difference (MCID) for pain and disability measures.
The GRADE analyses showed a moderate level of evidence on the short-term effects of Pilates compared to minimal intervention and no effect if compared with other types of exercise concerning pain (Table 4). For disability, the level of evidence was low for this comparison. For both pain and disability, the evidence was downgraded due to the low and moderate quality of most of the reviews. Moreover, an additional down-grading for disability was added, since the results were con icting concerning the conclusion on the effectiveness of Pilates over minimal interventions.

f) Sling exercises
Sling exercises use slings and elastic bands to offset body weight and progress the exercises without pain (73). We found two systematic reviews on the effect of sling exercises (37, 60) including one meta-analysis (37). In total, 16 studies were included, but there was a very high overlap between the two reviews (CCA 23%). The last updated search in the systematic review was August 2013. The methodological quality of the included studies according to PEDro assessments was moderate and low. The AMSTAR-2 ratings for these two studies was 8 (60) and 14 points (37), indicating low and high quality studies, respectively ( Table 1). The interventions in both reviews were primarily sling exercise based; however, the sling exercises were also combined with e.g. passive modalities and with other kind of exercises, back school, contemporary treatment, and drugs. The control groups received other forms of exercise, passive modalities, manipulation, contemporary treatment, and drugs (APPENDIX 4).
The narrative analyses of the reviews showed that sling exercises are no more effective in reducing pain or improving disability compared with other types of exercise (Table 3f). However, in comparison to passive modalities or the combination of physical agents and drug therapy, sling exercises were more effective in decreasing pain and improving disability, but only sling exercise vs thermomagnetic therapy showed clinically relevant differences between the groups in favour of sling exercise. In addition, sling exercises were found to be not more effective compared to traditional Chinese medical therapies (37). Sling exercises in addition to acupuncture therapy were as effective as acupuncture therapy alone for reduction of pain and improvement of disability.
The GRADE analyses showed that there is moderate and low level of evidence for short-term and long-term effects on pain and disability, respectively, for sling exercises over thermomagnetic therapy; however, the evidence is only based on two RCTs (Table 4). We downgraded due to lack of precision and, for disability, also for inconsistency of the results. g) Traditional Chinese Exercises (Tai chi/Qigong) Tai chi and Qigong, two common types of traditional Chinese mind-body techniques, also referred to as Traditional Chinese Exercises (TCE), include low to moderate intensity exercises coordinated with slow body movement and focus on physical-mental connection (38). Two reviews were identi ed evaluating the e cacy of Tai Chi and Qigong, and both performed a meta-analysis. Included were 10-11 RCT's published between 2008 and 2019, and the total sample size ranged between 886 and 959 participants. There was a high overlap of the studies investigating Traditional Chinese Exercises (CCA 13%). The studies included in the reviews were rated as having fair to good quality. The risk of bias was assessed to 11.5 and 13 out of 16 AMSTAR-2 points, indicating moderate to high quality (Table 1). Both reviews compared the e cacy of either Tai Chi or various types of Qigong (Wuqinxi, Baduanjin, Liuzijue) to either no treatment, active treatment (strength exercise, backwards walking, or other physiotherapy), or usual care, with or without the experimental component (APPENDIX 4).
The narrative synthesis on pain showed small to moderate effects for Traditional Chinese Exercises over no treatment, active treatment, or usual care only. Subgroup analyses revealed a larger effect when Tai Chi was compared to no treatment, than to active control interventions or to routine care (without an added Tai Chi component) (Table 3g). Only short-term effects seem to have been evaluated, but exact follow-up time was not reported. The synthesis on disability showed a variability in effect, from small to large effect for Traditional Chinese Exercises over no treatment, active treatment or usual care only (Table 3g). In both systematic reviews, the effects differed depending on outcome measure used (38, 49).
The GRADE analyses for pain showed a moderate level of evidence concerning pain and a low level of evidence for disability on the short-term effects of Traditional Chinese Exercises compared to no intervention (passive control), various active treatments or usual care in chronic LBP patients concerning pain ( Table 4). The evidence was downgraded for imprecision due to heterogeneity (pain, disability) and additionally due to large con dence intervals of the effects (disability).

h) Walking
Walking interventions use outdoor walking (with or without supervision), treadmill walking, and/or Nordic walking as therapeutic programs in chronic LBP (51). The literature search identi ed three systematic reviews (34, 43, 51) of the effectiveness of walking interventions, two of which performed a metaanalysis (34, 51). The reviews included ve to nine studies with 329 to 869 participants with a very high overlap of the studies (CAA 37.5%). The reviews were published between 2016 and 2019, with the last updated search up in October 2017 (34). The quality of the included studies was low (43) to high (34). Two reviews achieved 10.5 out of 16 AMSTAR-2 points, indicating moderate quality (43, 51), and one achieved 15 out of 16 AMSTAR-2 points, indicating high quality (34). Two reviews (43, 51) did not report excluded studies, the source of funding, and did not investigate the impact of study quality on summary estimates ( Table 1). All reviews compared the effectiveness of walking interventions (overland and/or treadmill, and/or Nordic walking) to nonpharmacological interventions (e.g. other types of exercise, physical therapy, education), and two additionally compared walking and exercise to exercise alone (34, 51) (APPENDIX 4).
Both the meta-analyses (34, 51) for either pain or disability and the SR (43) for disability found no signi cant differences between walking and the comparison groups that received other interventions (Table 3-H). Neither did adding walking to the comparison groups induce a statistical improvement.
The GRADE analysis showed that there is a low quality of evidence that walking is as effective as other non-pharmacological interventions for pain and disability improvement in chronic LBP patients, and adding walking to exercise does not increase effectiveness ( Table 4). The evidence was downgraded due to study limitations and for imprecision due to large con dence intervals of the effect, and a large overlap of the reviews.

i) Yoga
Yoga exercises follow the traditional Yoga principles with a physical component (41). We included nine systematic reviews on Yoga (20,33,41,53,54,58,59,63,68). Five out of nine conducted a meta-analysis (20,33,41,53,59). The reviews included between four to 14 RCTs and 403 to 1444 participants, with a very high overlap (CCA 28%). The publication year ranged from 2011 (63) to 2019 (53), and the last updated search was in 2018 (53). The included studies were graded with varied quality. The study quality ranged from 3 (68) to 16 points (20) on AMSTAR-2 (Table 1). Only two reviews were rated as having high quality (20,33). Only one study presented a list of the excluded studies (20). The Yoga interventions were highly heterogenous, not only in terms of which kind of Yoga was used, but also in the length, frequency, and intensity of the sessions. Some interventions were combined with other physical therapy modalities, with book readings or usual treatments. There were no clear manuals or protocols that described the Yoga interventions. The control interventions were treatment such as physical therapy, waitlist control, stabilizing exercise and physical therapy, conventional exercise therapy, usual care, educational control group, and self-directed medical care (APPENDIX 4).
The narrative synthesis on both pain intensity and disability in the included reviews showed a short-term effect for Yoga, especially compared to no or minimal intervention, but also compared to general exercises. Three MAs showed medium, and medium-to-large effects, indicating that the effects of Yoga may be of clinical importance (33, 41, 59). However, the long-term effects did not seem to demonstrate better effects than usual care (Table 3i).
The Grade analyses showed a low level of evidence for a short-term effect in pain and disability for Yoga over general exercises; however, the long-term effects did not seem to demonstrate better effects than usual care or compared to usual care or compared to other types of exercises (Table 4). We downgraded due to large heterogeneity between the studies and inconsistent results. Although the risk of bias was high in most of the reviews, two reviews had low risk of bias (16 points); hence we decided not to downgrade due to study limitations.

Discussion
We aimed to summarize and synthesise systematic reviews and meta-analyses of the effects on pain and disability of common exercise types used in chronic LBP treatment. All in all, we included 41 reviews on various exercise types in chronic LBP and based on data searches up until November 2018. We found low to moderate evidence that participating in any exercise type is effective for reducing pain and disability compared to no or minimal intervention, but that no exercise type is more effective than another (very low to moderate evidence).
The results from this review of reviews are mainly in keeping with previously published reviews on the effects of exercise in chronic LBP, adding to the existing evidence and that no type of exercise seems to have a better effect over another (8,14,23,24). Overall, exercises seem to have a better effect on pain and disability than minimal intervention or usual care, but show no difference compared to other non-pharmacological interventions. This seems to be true in both the short and long-term. A Cochrane review on exercises in several chronic musculoskeletal disorders concluded that physical activity and exercise is an intervention with few adverse events that may improve pain and physical function, and consequently quality of life (14). Geenen et al. (2017), further concluded that the quality of evidence on included studies is low, and that future studies should focus on increasing participant numbers and lengthening both the intervention itself and the follow-up period (14). In addition,  summarized in a Cochrane review including 61 studies, that exercise therapy seems to be slightly effective at decreasing pain and improving function in adults with chronic LBP, concurring with our study ndings (8). Five years later, Van Middelkoop et al. (2010) reviewed studies on exercises in chronic LBP also concluded that no evidence exists that one exercise type is more effective than others, and that it remains unclear which subgroups of patients bene t most from a speci c type of treatment (24). However, recently, a network analysis of the effectiveness of exercises in LBP suggested that there is low quality evidence that Pilates, MCE, resistance training and aerobic exercises are the most effective treatments (23). The authors compared various types of common exercises with no or minimal interventions and their results are partly in line with ours. In addition, our review included results on less studied exercise types such as aquatic training, walking, Traditional Chinese Exercises, and sling exercises, which all also showed consistency with previous results that no exercise type seems to be more bene cial than another.
Apparently, some exercise types prescribed in chronic LBP are more studied than others. There has obviously been a research focus on MCE, Pilates and Yoga, based on the number of published reviews we found in our data base searches (MCE n = 11; Pilates n = 9; Yoga n = 9) compared to the other exercise types included in our study. Eleven reviews showed mainly consistent results that MCE is as effective as manual therapy regarding pain and disability, and showed overall statistically signi cant but clinically not relevant ndings that MCE is more effective than general exercise. However, in comparison with minimal intervention, our results showed a clinically important difference. As most of the included reviews showed small effects for MCE in comparison with other interventions on pain and disability, we suggest -based on our ndings -that no future studies will change the current evidence. However, Saragiotti et al.
(2016) (18) suggested that there might be sub-groups in LBP for which MCE might be more bene cial. This might, of course, be the case regarding all exercise types. For Pilates, there was low to moderate evidence that it is no better than other exercises, however better it is than minimal interventions; small effect sizes were found. Pilates and MCE might be comparable as exercise types but differ in that MCE seems to be more often supervised and also performed as a graded program, starting with low load and speci c exercises. Hayden et al (74) concluded that exercise therapy that consists of individually designed programs, and is delivered with supervision may improve pain and function in chronic nonspeci c LBP. Moreover, adherence to exercise programs has been shown to be highly correlated with positive outcomes (74,75). We did not, however, summarize adherence to exercises or if a program was performed individually or in group. This might be of value for future reviews. Sling exercises is also an exercise type that aims to stabilize the spine and the results for this type of exercise are consistent with previous literature.
Yoga showed some effects compared to general exercises in the short and long-term. Yoga is an exercise that also includes breathing training and covers mental aspects, which probably affects several systems in the mind and body (20). As the included Yoga interventions were heterogenous in terms not only of what type of Yoga was used, but also in length, frequency and intensity of the sessions, speci c evidence is hard to establish.
For some exercise types, such as resistance training, we found few, and mainly older systematic reviews, perhaps because the trend in how to exercise in LBP has affected the number of studies. Three reviews published between 2010 and 2012 (66,67,69), showed that there is very low evidence that resistance training is no more effective than other interventions. To use and to investigate more loaded training in chronic LBP has seemingly not been on the research agenda over the last decade; however, recently, a new interest in studying loaded exercise in LBP seems to have appeared (76,77), perhaps as a reaction to two decades of study interest in low load exercises such as MCE.
For walking, aerobic, aquatic, and Traditional Chinese Exercises we found only few reviews. Walking was investigated in three systematic reviews and showed a moderate certainty evidence of no effect on pain and disability compared to other exercises or physiotherapy. This was also the case with aerobic exercises only investigated in one review (36). For aquatic exercises, the only review showed a pain reduction in the intervention group compared to the comparison group performing land-based or no exercises, but with low quality evidence and questionable clinical relevance (50). In Chinese Traditional Exercises, two MAs showed that these are more effective than no treatment, but only short-term effects were found. In our review, we chose to analyse these exercise types, while other reviews have compiled them into wider categories, i.e. walking could also be seen as an aerobe exercise intervention (23). We believe that in future research, one should keep these exercise types separate in order to be able to establish the level of evidence for speci c exercise types.
To date, exercises are suggested by most clinical guidelines in chronic LBP and commonly used, but with no clear recommendation for one type of exercise over another (6,(78)(79)(80). Instead, it is recommended that exercise be prescribed based on patient preference and clinician experience (6). This broad recommendation probably re ects uncertainty about the mechanism(s) through which exercise yields positive effects on pain and function (21). It could also be that physiotherapists prescribing exercise programs have different preferences for speci c exercise types based on education and interest, or nonawareness of clinical guidelines (81).
To decide on what exercise should be chosen for an individual patient suffering from chronic LBP, the patient should be included in the decision-making process. Gardner et al (82) propose better goal setting in chronic LBP patients to determine treatment interventions that are driven by patient preference. Thus, our research might increase the current knowledge of effectiveness of various exercise types.

Strengths and limitations
A strength of our study is that, to our knowledge, this is the rst review of reviews on various types of exercise used for treatment of chronic LBP, including exercise types that have not been studied in this way previously. We did not limit our search but included reviews in all languages without any restrictions on publishing year or comparators. Furthermore, we followed and complied with the PRISMA guidelines, and graded the study quality using the recommended instrument AMSTAR-2 (30), and summarized the graded evidence of the different exercise types. Two-thirds of the included reviews were of moderate to high quality, and 12 studies were assessed as having a critically low or low quality. Still, the evidence for the included reviews was mostly downgraded due to study limitations, and further high-quality reviews are therefore warranted. Our study is a collaboration of 10 researchers, which brings both advantages and disadvantages. We worked in pairs to include the reviews, to draw the data to the tables, and to assess the RoB. A third arbitrary party was always used when no consensus could be reached. This procedure was necessary in order to manage the enormous number of studies, and of course could have led to differences in how the various exercise types were judged. However, wesolved this by changing the pairs so that in all the processes more or less four researchers were involved in the studying and extracting of data for each exercise type. There are other limitations that need to be discussed, such as inclusion bias, since we might have missed some important reviews to include in our database search. However, the data basesearches were conducted using relevant search strategies in several databases by a librarian from the Karolinska Institutet. Moreover, the reference lists of the included reviews were studied for additional reviews to include. A publication bias might be that we did not perform any further search of the grey literature on, for example, web pages. We still consider that we found most relevant literature in our searches. We thus considered that additional reviews would probably not have changed our main ndings to a large extent.
The results are based on several reviews and meta-analyses with a high or very high overlap of included studies. Overlap could be a problem if one study that is included in all reviews drives the results in the review in one direction. This could have been the case in the reviews of Yoga and Pilates, where one highqualitative study with bene cial results was included in all reviews and might have affected the overall results (83, 84). It might be discussed whether a review of reviews is the best way to summarize the evidence as the results from a review of reviews summarize and mirror the risk of bias in all of the included reviews. Might it be more relevant to include all studies in one large review instead, to cope with problems such as overlap? Such a network review on chronic low back pain was recently published by Owen et al. (2020), including 89 studies. However, to conduct a review of review is a more feasible way to, in one large publication, summarize the existing evidence on the effectiveness of various and speci c exercise types used and prescribed in chronic LBP.

Clinical and future perspectives
Exercise is recommended in most guidelines (6,11). Our study ndings showed that all nine exercise types could be as useful as each other in rehabilitation of chronic low back pain. Since no exercise type shows any better effect than any other the choice of exercise type should therefore be based on patient preference. Patient preference has been reported to have an even better effect on pain and disability measures than following guidelines (85). In this review, the primary outcomes studied were pain and disability. Further studies should therefore incorporate other outcomes that re ect additional effects of exercise, e.g. fear-avoidance, quality of life, pain catastrophizing. Moreover, the choice of outcome should also be more speci cally related to the goals of the exercises. For some exercise types, such as aquatic and sling exercises, more high-quality research is warranted, but for others, such as MCE there is a need for speci cally analyzing different sub-groups.

Conclusion
Looking at exercise from a broad perspective, we found low to moderate evidence that participating in any exercise type is effective for reducing pain and disability compared to no or minimal intervention, but that no exercise type is more effective than another (very low to moderate evidence).

Functional ability increased in 7 trials
Evidence suggests that RT can increase muscle strength, reduce pain and improve functional ability in patients suffering from CLBP, RT can be used successfully as a therapeutic modality in several musculoskeletal conditions, especially those of a chronic variety. Although the exact application of training intensity and volume for maximal therapeutic effects is still unclear, it appears that RT guidelines, which have proven effective in a healthy population, can also be successfully applied in a rehabilitation context. There is moderate evidence that a combination of endurance training and progressive resistance training of the back muscles is more effective than no intervention, but equal effective as a cognitive behavioral intervention. Future high quality RCT's will have to clarify whether MTT is effective and would be superior to other forms of therapeutic exercise. In comparison with passive treatment or no treatment, there is strong evidence for the bene t of resistance training, but nonspeci c tness training is comparable effective in rehabilitation.
Abbreviations: CBI=Cognitive Behaviour Intervention; CLBP=Chronic Low Back Pain; MA=Meta-analysis; NR=Not Reported; MTT=Minimal intervention; RCT=Randomized Clinical Trial; RT= Resistance training; SR=Systematic review. Pilates based therapeutic exercise was found to be moderately superior to minimal intervention for pain relief and confers similar bene ts when compared with pooled scores to another physiotherapeutic treatment but should be interpreted with caution.
Pilates is moderately better than another physiotherapeutic treatment in reducing disability and provides comparable bene ts to minimal intervention.
Future studies should incorporate placebo-controlled trial, larger sample sizes, intervention protocols that are comparable, assessment of the several features not coded in this review and longer-term follow-up. Pilates is superior to minimal intervention for reduction of pain.
Pilates is not more effective than other forms of exercise to reduce pain. Pilates is no more effective than minimal intervention or other exercise interventions to reduce disability. There is a need for well-designed randomized controlled trials with adequate follow-up. Pilates offers greater improvements in pain and functional ability compared to usual care and physical activity in the short-term. Changes in pain are more likely to be clinically signi cant than improvements in functional ability.
Pilates offers equivalent improvements to massage therapy and other forms of exercise. Future research should explore optimal Pilates designs, and whether some people with CLBP may bene t from Pilates more than others. Yamato  No high-quality evidence for any of the treatment comparisons, outcomes or follow-up periods investigated. Low to moderate quality evidence that Pilates is more effective than minimal intervention for pain and disability. When Pilates was compared with other exercise we found a small effect for function at intermediate-term follow-up. Thus, while there is some evidence for the effectiveness of Pilates for low back pain, there is no conclusive evidence that it is superior to other forms of exercises. The decision to use Pilates for low back pain may be based on the patient's or care provider's preferences, and costs.     Pain and disability were similarly improved by walking or exercise, no additional improvement when walking is added to exercise.
The low clinical relevance of the outcome was not su cient to make recommendations.
Abbreviations: MD=Mean difference; LBPFS=Low back Pain Functional Score (0-100); LBPFS=Low back Pain Rating Score; MA=Meta-analysis; NRS=Numerical Rating Scale (0-10), ODI=Oswestry Disability Index (0-100); RMDQ=Roland-Morris Disability Questionnaire (0-100); SR=Systematic review; VAS=Visual Analogue Scale (0-100). Yoga appears as effective as other non-pharmacologic treatments in reducing the functional disability of back pain. It appears to be more effective in reducing pain severity or "bothersomeness" of CLBP when compared to usual care or no care. Yoga may have a positive effect on depression and other psychological co-morbidities, with maintenance of serum BDNF and serotonin levels. Yoga appears to be an effective and safe intervention for chronic low back pain. Yoga may represent an e cacious adjunctive treatment for CLBP; the effect size for yoga in reducing pain and functional disability appears to be similar to, if not higher than, effects sizes for more traditional exercise therapy, cognitive behavioural therapy and acupuncture). Overall, the ndings provide the strongest support for the effects of yoga on short-term improvements in functional disability among patients with CLBP; a range of different yoga interventions yielded statistically similar effect sizes. Five RCTs suggested that yoga leads to a signi cantly greater reduction in low back pain than usual care, education or conventional therapeutic exercises. Two RCTs showed no between-group differences. It is concluded that yoga has the potential to alleviate low back pain. However, any de nitive claims should be treated with caution. (2 trials).
Yoga was found to be more effective for pain at three and six months; uncertain whether there is any difference between yoga and other exercise for back-related function or pain, or whether yoga added to exercise is more effective than exercise alone.
There is a need for additional high-quality research to improve con dence in estimates of effect, to evaluate long-term outcomes, and to provide additional information on comparisons between yoga and other exercise for chronic non-speci c low back pain. Yoga, may be bene cial for CLBP symptomatic management, irrespective of non-control comparison or active control comparison (conventional exercises, core training, and physical therapy programs). There were also few Tai