Exercise interventions can improve muscle strength, endurance, and electrical activity of lumbar extensors in individuals with non-specific low back pain: a systematic review with meta-analysis

Exercise interventions have been recommended for people with non-specific low back pain. The literature is scarce regarding the effects of exercise on muscle strength, endurance, and electrical activity of lumbar extensor muscles. Electronic searches were carried out from May 2020 until August 2020 in the following databases: PUBMED, CENTRAL, EMBASE, PEDro, SPORTDiscus, Scielo, and LILACS. Only randomized controlled trials with passive and active control groups were included. The methodological quality of the included studies was performed using the Physiotherapy Evidence Database Scale. Eight studies, involving 508 participants, were included in metanalytical procedures. Exercise interventions demonstrated superior effects on muscle activity (Electromyography) when compared with active controls (p < 0.0001). Exercise interventions demonstrated superior effects on muscle endurance (Sorensen Test) when compared with passive (p = 0.0340) and active controls (p = 0.0276). Exercise interventions demonstrated superior effects on muscle strength (Machine) when compared with passive controls (p = 0.0092). Exercise interventions can improve muscle strength, endurance, and electrical activity in people with non-specific low back pain.

Approximately 80% of adults experience lower back pain (LBP) at some time in their lives 1 . In 1990, for all ages and both sexes, the leading cause of years lived with disabilities was LBP (42.5 million years lived with disabilities) 2 . Between 1990 and 2007, the number of all-age years lived with disabilities attributed to LBP increased by 30% 2 . LBP leads the cause of years lived with disabilities in 126 of 195 countries according to the Global Burden of Disease Study from 2007 to 2017 2 . In 2019, for ages 50-74 years, LBP remained in the top-ten-ranking causes of years lived with disabilities 3 .
LBP has a strong socioeconomic impact 4 , which leads to a decrease in people's quality of life and productivity 5 , and an increase in direct and indirect costs with palliative treatments 6 . The annual cost of a person with LBP is approximately 7000 euros 7 . When added to the unproductive occupational behavior, the costs rise to approximately 18,000 euros 7 . In Brazil, the annual loss of productivity per individual with LBP costs approximately 2684 Dollars 8 , and in the US the annual loss of productivity cost per individual is approximately 1685 Dollars 9 . LBP also affects employment in the informal sector 10 , which raises the hypothesis that the data mentioned above could be higher.

Inclusion criteria. Types of studies.
We included only randomized controlled trial studies 21 performed for more than 6 weeks. In the preliminary searches, a sufficient number of randomized controlled trials were found to justify this criteria design for eligibility and answer the study question 22 . Therefore, we decided not to include non-randomized controlled trials 22 . Regarding the six week criteria, this period was chosen based on previous literature, according to the initial phase of neuromuscular adaptations from resistance training 23 .
Type of participants. We included studies with any type of NSLBP (acute, sub-acute, chronic) in adult individuals, aged between 18 and 55 years, with no restriction on sex. LBP was defined as pain and or discomfort located below the ribs and above the gluteal crease 24 . NSLBP is not attributed to a recognizable or specific pathology 25 and we considered for this study LBP with or without referred leg pain. We excluded studies with participants that had undergone spine surgery, osteoporosis, fractures, and malignancies. Patients with systematic diseases or non-mechanical LBP (e.g., disc herniation, spinal stenosis, etc.), who were pregnant, experienced postnatalrelated LBP, and military forces were excluded.

Type of interventions.
We included studies comparing an experimental group (exercise interventions for increasing trunk extensor muscle function) versus passive and or active controls. For passive controls, we considered: no intervention and waiting list groups. For active groups, we considered: standard care (e.g., multimodal physical therapy) and different types of exercise that are not explicitly designed to increase muscle strength, such as aerobic exercise, Yoga, stretching exercises, home-based exercises, circuit-based exercises, telerehabilitation, and Tai Chi Chuan. We followed the Cochrane Handbook for Systematic Reviews of Intervention to define the control group classifications 22 . When the experimental group was used in addition to another active treatment, the trial was included (e.g. [exercise intervention plus stretching] versus [stretching]) . We excluded studies that compared two different types of exercise interventions for increasing muscle strength, endurance, or electrical activity of trunk extensors (e.g., motor control exercise versus machine strength exercise). This decision was made considering that there is no standard gold method of exercise for the treatment of LBP patients 15 and the present study was not developed to investigate the best exercise (comparisons between exercises designed to increase muscle strength for example).
We decided to cluster the analysis of interventions (different exercises), considering that they fit the definition of physical training, in which the muscle moves or tries to move against an opposing force. In the case of isometric exercises, we considered that gravity is a force to be overcome 26 .
Types of outcome measures. Continuous data for meta-analysis were obtained from general outcomes designed to assess 21 muscle strength, muscle endurance, and muscle activity of trunk extensor muscles 24 , such as EMG, and muscle strength measured using direct (e.g., isometric and dynamic dynamometers) and indirect (e.g., Biering-Sorensen test) methods. www.nature.com/scientificreports/ To analyze muscle activity and fiber recruitment, surface electromyography equipment (time and frequency analysis) has been employed as the gold standard for many years to study normal and altered outcomes, such as maximal isometric muscle contraction, and the units are presented in Hertz 27 . The isokinetic dynamometer allows assessment of strength during a dynamic or isometric contraction. The dynamometer resistance is equal to the muscular forces applied to the machine, and the units presented are in Newton-metres 28 . The Sorensen  test measures the amount of time a person can hold the unsupported upper body in a horizontal prone position  with the lower body fixed on the examining table. The units presented are in s 29 . Search methods for identification of studies. Electronic searches started in May 2020 and were conducted in the following databases until August 2020: PUBMED, CENTRAL, EMBASE, PEDro, SPORTDiscus, Scielo, LILACS. Only articles written in English were included, but there were no restrictions imposed on the publication date. The "ClinicalTrials.gov" database was used to identify potential unpublished studies and ongoing studies. Google Scholar was used to assess the grey-literature (thesis, clinical report, conference abstract).
Research strategies were conducted and designed depending on the specific settings of each database. A dedicated search strategy was prepared for each database. According to the PICO model of a clinical question (only participants and interventions), MeSH (Medical Subject Headings) terms and text words (e.g., low back pain; exercise; strength training) were used and combined with Boolean operators (AND, OR). Additionally, a manual search was conducted through the bibliographies of all included studies to obtain an integrative crossreferenced full-text selection. We report the primary core search strategy used in the databases consulted (Supplementary Material, Supplementary Table S1). In addition, Endnote version 8.0 was used to assess duplicated references from the database searches.
Data collection and analysis. Selection of studies. Two review authors (SC, LFC) independently screened all titles and abstracts retrieved by the search strategy for eligibility. Those deemed potentially relevant were retrieved for full-text assessment by the same authors (SC, LFC), who assessed whether the reports fulfilled the selection criteria. When necessary, a third review author (WRM) resolved any disagreements regarding study inclusion. We used a PRISMA flowchart to summarize the search results and the study selection process 30 .
Data extraction and management. Two review authors (SC, LFC) independently extracted the primary data from the studies using a standard data extraction form on Excel software to collect the following details: participants, intervention, comparator, outcomes, assessment, conclusion, and financial support (Table 1). In addition, participants, intervention, comparator, and outcomes were extracted, as shown in Table 2. The extraction was checked by a third reviewer (ALAR).
Methodological quality. Two review authors (KLC, ALAR) independently assessed the methodological quality of the included RCTs using the Physiotherapy Evidence Database Scale (PEDro) scores. The PEDro scale consists of 11 criteria: random allocation, concealed allocation, baseline comparability, blind subjects, blind therapists, blind assessors, adequate follow-up, intention-to-treat-analysis, between group comparisons, point estimates and variability). The items assessed receive either a "yes", or "no" rating. The maximum PEDro score is 10 points. Trials with a PEDro score ≥ 6 points were classified as high-quality, while trials with a PEDro score of < 6 were classified as low-quality 31 . Any disagreement was resolved by a third review author (WRM).
Measures of treatment effects. Considering that the values of outcomes investigated were continuous variables and the scale of measurement, the mean differences (MD) and 95% confidence intervals (CIs) were used. The MD can be used as a summary statistic in a meta-analysis when all study outcome measurements are made on the same scale 22 . The MD is a standard statistic that measures the absolute difference between the mean values in the groups of a randomized trial. A common practical problem in the meta-analysis of change scores is when the study did not report the standard deviation (SD) of change scores; therefore, we decided to extract the data from post-intervention values (this assumption avoids the need to impute the SD of the changes) 22 . The postintervention values for meta-analysis procedures were obtained using the first time point close to the end of the treatment because few studies reported follow up measurements. For statistical analysis, the continuous data were extracted to a database on Excel Software (Version 16.42) before using RStudio software (Version 1.4.1106, RStudio, Inc) with the following packages: "meta", "metafor", "readr", "Rcpp", "BH" and "readxl" to perform the appropriate metanalytical procedures.
Assessment of heterogeneity and sensitivity. The heterogeneity of the studies was assessed by the I 2 statistic and 95% CI 32 . The following I 2 statistics were considered: 0-40% might not be significant, 30-60% may represent moderate heterogeneity, 50-90% may represent substantial heterogeneity, and 75-100% may represent considerable heterogeneity 32 . Since the included studies have distinct populations, intervention parameters, and settings, a random-effect was always used. This decision was made based on the expectation that the intervention effects are not truly identical between studies. We decided not to choose between fixed-effects and random-effects according to the statistical test results for heterogeneity 22 . Considering that the variables used to perform the meta-analytical procedures were established clearly and a priori (eligibility criteria, continuous data [analysis on post-intervention], and analysis methods [random effects; mean difference dimension]), the sensitivity analysis was not employed considering these assumptions. www.nature.com/scientificreports/ Level of confidence in meta-analytical results. The quality of the evidence was rated using the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE). GRADE offers four levels of evidence: high, moderate, low, and very low. Randomized trials begin as high quality evidence, and the quality may be downgraded according to limitations in five domains: study design and risk of bias, inconsistency of results, indirectness of evidence, imprecision, and other (for example, publication bias). If there were sufficient data available to use quantitative analysis for summarising the data, we assessed the quality of the evidence for each outcome. To summarize the rating of the quality of evidence to make recommendations, the GRADE pro system was used for each outcome (https:// grade pro. org/) 33 . Thus, we also presented the results using the summary of findings tables. In the subgroup analysis, two GRADE assessments were performed (one for each subgroup). www.nature.com/scientificreports/ Clinical relevance. Assessment of clinical relevance was carried out using three categories: small effect (MD < 10% of the scale; SMD < 0.5); moderate effect (MD from 10 to 20% of the scale; SMD from 0.5 to 0.8); large effect (MD > 20% of the scale; SMD > 0.8) 34 .

Results
The electronic search retrieved 14,389 documents, of which 12,793 were excluded as duplicates, 1464 were excluded after screening by title and abstract, and 18 were excluded after full-text reading. Therefore, 17 studies [35][36][37][38][39][40][41][42][43][44][45][46][47][48][49][50][51] were included in the qualitative synthesis after applying the eligibility criteria. Of these, six were included in the meta-analysis 35,37,41,46,48,51 . Figure 1 shows the search phases and screening of the studies included in the qualitative (systematic review) and quantitative (meta-analysis) synthesis.  Table 1. The ongoing studies identified in the clinical trial database are presented in Table 2. Table S2 shows that the PEDro score ranged from  42 did not report data appropriately (the data were presented in graphs; strength values were adjusted using body weight;  49 and Lomond et al. 44 were unable to provide the data and the other authors did not respond to the email.  www.nature.com/scientificreports/ of two studies). The clinical relevance found was small (Δ 8.42%). There was no heterogeneity in the muscle electrical activity analysis between exercise vs. active control on the EMG (I 2 = 0%; p = 0.83).

Meta-analysis.
The meta-analysis on muscle endurance of trunk extensors demonstrated statistical difference in favor of exercise interventions when compared to passive control ( Fig. 3; n = 37 participants [experimental n = 20; control n = 17 participants], MD = 44.27 s [3.33, 85.21], p = 0.0340), with very low confidence in the effect estimate (Fig. 6, GRADE analysis of two studies). Large clinical relevance (Δ 31.39%) was found. There was substantial heterogeneity between exercise vs. passive control in the analysis of muscle endurance of trunk extensors (I 2 = 73.17%; p = 0.05).
The meta-analysis on trunk extensor muscle endurance demonstrated statistical difference in favor of exercise interventions when compared to active control ( Fig. 4; n = 105 participants [experimental n = 50; control n = 55 participants], MD = 21.99 s [2.43, 41.56], p = 0.0276), with low confidence in the effect estimate (Fig. 6, GRADE analysis of two studies). Moderate clinical relevance (Δ 11.01%) was found. There was no heterogeneity between exercise vs. active control in muscle endurance of trunk extensors analysis (I 2 = 0%; p = 0.78).
The meta-analysis on muscle strength demonstrated statistical difference when compared to passive control (    The exercise interventions performed in the studies included in this systematic review are all classified as resistance training exercises. The literature review studies on this topic (not restricted to LBP) have presented exponential growth in recent decades, with more than 552 systematic reviews with meta-analysis published regarding resistance training exercises in the PubMed database. The classic outcomes in many of these reviews are changes in strength and hypertrophy, under different conditions, after manipulating acute and chronic training variables [52][53][54][55][56][57] . Resistance training is already recognized internationally as a medicine 58 , which is recommended in various conditions and diseases 59 . In the present review, we assessed three outcomes related to trunk extensor muscle function: strength, endurance, and myoelectrical activity.
The results with major clinical relevance were the effects of exercise interventions on muscle endurance when compared to passive control (large effect), with the control being less effective, although it is worth mentioning that few subjects were included in this analysis. There are multiple risk factors for developing back pain, including low back extensor endurance, and identifying these potential risks may be important in clinical practice 60 . Trunk extensor muscles are designed to support continuous activity throughout the day, but pain and inactivity alter these muscles so that they fatigue during activities of daily living 61 . The effects of exercise training in the present study were demonstrated by studies that used the Sorensen test, probably the most clinically useful test for clinical practice settings 62 . A previous study demonstrated that patients with chronic low back pain presented lower back extensor muscle isometric endurance than healthy subjects during the Sorensen test 63 . Here, the back muscle endurance outcome demonstrated that exercise interventions could be emphasized in rehabilitation strategies for subjects with chronic and subacute NSLBP. On the other hand, failure to exercise can increase general chronic pain 64 and subacute LBP 65 .
The Sorensen test has also been used to analyze the trunk extensor fatigability based on the median frequency of electromyography analysis, and patients with LBP presented a significantly lower median EMG frequency in thoracic and lumbar regions, suggesting that individuals with low back pain demonstrated higher trunk fatigability 63,66 . Although in the present study the EMG of lumbar extensors demonstrated a small clinical effect, the results of the exercise intervention were superior to active control groups. Thus, the exercise interventions could also be indicated for subjects with NSLBP using some traditional modalities (multimodal physical therapy and treadmill walking exercise). Previous evidence from numerous studies demonstrated that lumbar extensors are active (EMG) during the performance of various exercises resulting from acute training 17 . Therefore, there is now some evidence of the potential of chronic adaptations using exercise training on machines for trunk extensors. Exercise interventions increase motor unit recruitment and firing rate 67 , and these alterations can increase muscle endurance 23 . In addition, resistance training increases EMG amplitude and muscle strength, suggesting . c = The studies were classified as low quality (score 3/10 = Cortell-Tormo et al. 72 ; score 4/10 = Kell et al. 51 ). d = The heterogeneity between studies were substantial (I2 = 73%; p = 0.05). e = The confidence interval are very large (3.33-85.21 s). f = The studies were classified as low quality (score 5/10 = Mannion et al. 46 ; score 4/10 = Kell et al. 51 ). g = Bruce-low et al. 37 . The study does not provide concealed allocation and blinded assessments (score = 5/10). h = The heterogeneity between studies were reported by I2 statistics as 58% (substantial heterogeneity www.nature.com/scientificreports/ a neural contribution 23 . Motor neurons, known as the final common pathway of neural activation signals, are improved by resistance training, leading to upregulation of agonist activity and possible intermuscular coordination of synergist muscles 68 . Likewise, resistance training can improve neural adaptations. Resistance exercise training also improves mitochondrial size and quantities 69 , which leads to an increase in ATP production 23 . Furthermore, mitochondria are responsible for lactate oxidization, which transforms lactate into glucose and provides body energy through the Cori cycle 23 . These statements show that resistance training can improve endurance, and people with NSLBP should use resistance training to improve trunk extensor muscle endurance. Regarding muscle strength (exercise interventions vs. passive control), only a small clinical effect was demonstrated. These were surprising results because resistance training exercise has collectively been shown to be effective in increasing strength compared to non-exercise training-based treatments in adults 26 . Muscle weakness can lead to increased pain 70 and decreased functionality 71 , and strength training is considered a treatment for these situations 72 . The dose-response to obtain gains from resistance training is a minimum of 4 sets per muscle group per week 73 . Neither of the studies used in the meta-analysis met this recommendation. One study 37 performed only two sets per muscle group per week, and the other study 41 performed only 1 set per muscle group per week. It is believed that the small clinical effect is due to not using the dose-response reported in the literature.
Other systematic reviews show that exercise interventions are effective and safe 15 on subjective outcomes, such as reducing pain 13,14 , functional limitations 13,16 , disabilit 14 , and time to return to work 16 . In addition, strength training stimulates the release of serotonin and endorphins in the brain, which reduces pain and improves mood 74 . Therefore, our meta-analysis is in accordance with the positive results of previous systematic reviews [13][14][15][16] that employed patient-centered outcomes (questionnaires). This means that exercise interventions also improve objective outcomes, such as muscle strength, muscle endurance, and electrical muscle activity. For practical and clinical application, exercise interventions, preferably resistance training, could be recommended for people with NSLBP.
This study has some limitations: First, the publication bias analysis was not employed considering the reduced number of articles included in the meta-analysis, such as analyzing the visual inspection (funnel plot) and the Egger test. These analyses require a minimum of 10 studies, according to the Cochrane Handbook. However, we performed a comprehensive search in many databases, and searches were also carried out in the gray literature and randomized clinical trial register databases. Second, despite the clinical effectiveness of exercise interventions, it should be noted that according to the GRADE analysis, there was no outcome with a moderate quality of evidence. The analyses show very low (muscle strength and endurance [passive control]) and low (electromyography and endurance [active control]) quality evidence that exercise interventions are effective when compared to the control groups investigated. Third, the influence of variables related to the exercise prescription (duration, frequency, number of repetitions, intensity, movement speed, and rest interval) 75 was not considered in the metaanalytical procedures. Although this influence can be analyzed by the meta-regression approach, unfortunately, the analysis could not be performed with only two studies. Fourth, we cluster all interventions, even with different exercises, despite the fact that there are different demands on physical capabilities for each exercise. Fifth, the instruments for assessing strength and endurance are different between the studies included in the meta-analysis. This is a common situation when combining studies for meta-analytical procedures. However, we standardized the measurement units to use the mean difference summary effect as a statistical approach, in order to provide clinical applicability to the results. Finally, although the meta-analysis procedures were performed with two studies for all outcomes, it was decided to maintain the meta-analytical 22 results to provide absolute values that could be extrapolated for health professional use. Future systematic reviews with meta-analysis are needed using studies with high confidence and filling the remaining gaps.

Conclusion
Our study demonstrates that chronic exercise interventions (more than 6 weeks) can be effective in adults with NSLBP and should be incorporated into clinical practice to promote muscle adaptations. There are few studies included in the meta-analysis (only 2 per outcome), and therefore the results should be taken with precaution. From the GRADE analysis, almost all included studies were of low-quality confidence, also the results show small clinical evidence for several of the outcomes. There was very low-quality evidence that exercise interventions were effective to increase muscle strength and endurance when compared to passive control (no intervention). There was low quality evidence that exercise interventions were effective to increase muscle endurance and myoelectrical activity when compared to active control (multimodal physical therapy, aerobic training, and treadmill walking exercise). www.nature.com/scientificreports/