Exercises for the prevention of non‐specific low back pain

Objectives

This is a protocol for a Cochrane Review (intervention). The objectives are as follows:

To evaluate the effects of exercise alone or exercise plus education compared with inactive control or education alone to prevent non‐specific LBP.

Background

Low back pain (LBP) is a global public health problem. Since the mid‐1990s, the age‐standardised prevalence and years lived with disability have both fallen by over 10%; however, LBP remains the leading cause of disability globally (GBD 2023). A major contributor to disease burden associated with LBP is its recurrent nature.

Although acute LBP has a favourable natural course, recurrence is extremely common (da Silva 2017; Hancock 2015). The best estimates suggest that between 30% and 70% of people with LBP will have a recurrence within one year after recovering from a previous episode (da Silva 2017; da Silva 2019; Machado 2017; Medeiros 2022). Exercises alone or combined with education used for preventing LBP, reducing recovery time, disability, and recurrence of new episodes (de Campos 2021; Huang 2020; Shiri 2017; Steffens 2016). Therefore, preventing both new episodes and recurrence of LBP is crucial not only to reduce the burden of pain and disability but also to avoid the need for treatment and the burden of healthcare costs (Foster 2018).

Description of the condition

LBP is a musculoskeletal condition defined as pain or discomfort localised below the costal margin and above the inferior gluteal folds, with or without pain referred down the leg(s) (Dionne 2008; Maher 2017). In people presenting to primary care with LBP, only 1% have a serious disease (e.g. cancer, infection, fracture, or inflammatory process), and about 5% present with radicular pain/radiculopathy or spinal canal stenosis (Hartvigsen 2018; Manusov 2012; Traeger 2017). Thus, most people with LBP are classified as having non‐specific pain (Bardin 2017; Manusov 2012). Non‐specific LBP can arise from many of the pain‐sensitive structures in the lumbar spine (e.g. intervertebral disc, facet joint, sacroiliac joint); however, there are no diagnostic tests available in primary care that can reliably identify the specific structures that are the source of the LBP (Han 2023). Non‐specific LBP is considered a lifelong condition, characterised by periods of recovery, recurrence, and continuing or fluctuating pain rather than single episodes (Dunn 2013; Kongsted 2016; Soares Oliveira 2021).

LBP is common and impacts both individuals and society. It is estimated that around 50% to 80% of people will experience LBP at some point in their lives (Gonzalez 2021; Hartvigsen 2018; Maher 2017). One systematic review of 165 prevalence studies from 54 countries estimated that the mean point prevalence was 11.9% and the one‐month prevalence was 23.2% (Hoy 2010; Hoy 2012). Prevalence was higher among females and people aged between 40 and 80 years (Hoy 2012; Wu 2020). LBP is also a major contributor to work absenteeism and is responsible for a large proportion of direct and indirect healthcare costs worldwide (Becker 2010; Carregaro 2019).

There are several ways to think about the prevention of non‐specific LBP. It is common to think about the prevention of a future episode in a cohort of people who are pain‐free at baseline. However, this way of thinking is difficult to apply to those who have persistent LBP, which is now recognised to be very common. In these people, one goal of management may be to prevent an exacerbation or flare‐up in their symptoms. In this review, we will include both these ways of conceptualising prevention of non‐specific LBP.

There is limited research in the field of LBP prevention (Foster 2018). Different strategies have been proposed for the prevention of first‐ever episodes or recurrence of LBP, including physical and behavioural interventions, ergonomic changes, and the reduction of physical work demands (Roman‐Liu 2020). Systematic reviews have shown that interventions such as education alone, lifting techniques, backpack weight control, ergonomic furniture, and shoe insoles are not effective in preventing LBP (de Campos 2021; Linton 2001; Michaleff 2014; Sahar 2007; Steffens 2016; van Duijvenbode 2008). In contrast, more current evidence suggests that exercise programmes alone or in association with education may be effective interventions to reduce the future impact of LBP (de Campos 2021; Sowah 2018; Steffens 2016). However, the limited number of primary studies and methodological limitations of current trials reduce confidence in these findings (Ferreira 2021a; Ferreira 2021b).

Description of the intervention

As a subcategory of physical activity, exercise is a planned, structured, repetitive, and purposeful activity to improve or maintain one or more components of physical fitness, performance, or health (CDC 2018; Dasso 2019; Garber 2011; WHO 2020). It can be categorised by type (i.e. aerobic, strengthening, flexibility, balance, neuromotor, and multicomponent) and by intensity (i.e. light, moderate, or vigorous) (CDC 2018; Garber 2011; WHO 2020). Exercise programmes for LBP prevention generally include specific strengthening of the trunk muscles, lower limb strengthening, stretching, and aerobic training (Bigos 2009; Choi 2010; Larsen 2002; Linton 2001; Moore 2012; Suni 2013). Currently, there is no consensus on the characteristics of the exercises that should be prescribed, such as type, intensity, frequency, and duration.

Educational strategies are commonly used in chronic health conditions to promote behavioural changes by increasing the patient's knowledge of their condition (Louw 2018). LBP education can include information on anatomical and biomechanical aspects of the spine, information about the disease, postural and ergonomic aspects, stress management, neurophysiology of pain, and others (Bardin 2017; Louw 2018). Some randomised controlled trials (RCTs) in the prevention field have shown that education combined with exercises reduces the risk of an episode of LBP at one year (Larsen 2002; Warming 2008).

Previous systematic reviews have investigated exercise therapy for preventing LBP. Exercise programmes were shown to be effective in reducing the number of recurrences of back pain in a 2010 Cochrane review (Choi 2010). More recently, other systematic reviews found that exercise alone and exercise combined with education have the potential to reduce the risk of future LBP episodes (Huang 2020; Steffens 2016). The prescription of an exercise programme can reduce the risk of a new LBP episode by 35%, and when associated with educational actions, this protective factor can increase to 45% (Steffens 2016). In addition, evidence suggests that exercise can decrease healthcare utilisation by about 11% to 22% and reduce the number of days off work associated with this condition by around 35% to 58%. (Chaléat‐Valayer 2016; Suni 2013; Wright 2005). Furthermore, engaging in an exercise programme has been associated with a lower risk of developing LBP (Shiri 2017).

How the intervention might work

Exercise is considered a complex intervention that influences individuals' physical, psychological, and quality of life (Geneen 2017). It is broadly recommended to prevent the number and severity of episodes of LBP (Choi 2010; Huang 2020; Shiri 2017; Steffens 2016). However, the mechanisms through which exercise achieves its effects remain unclear (Wessels 2007; Wun 2021). In recent decades, research has described many theoretical models to explain how an exercise programme could prevent first‐episode or recurrent LBP. Considering LBP as a complex and multifactorial condition, a unique model might not be able to explain or justify the effect of exercise training on this condition.

Some approaches are based on motor control theory, posture concepts, and movement impairment syndromes (Hides 2019). In general, they linked posture misalignment, poor trunk motor control, less muscle capacity to external demand, and joint relative stiffness with a higher risk of developing LBP (Hides 2019; Saragiotto 2016). Briefly, these models are based on the assumption that for normal postural control in the trunk region, it is essential that the muscles in this region have strength, endurance, and flexibility to meet the demands imposed by movement (Hodges 2003; Holm 2002; Shumway‐Cook 2000).

Neurophysiological mechanisms could also explain how exercise is beneficial in preventing LBP. Studies have shown that an exercise programme could promote hypoalgesia by activating the endogenous pain regulation mechanism (Koltyn 2014; Lewis 2012; Lima 2017). Moreover, exercises could have a positive effect on neuroplasticity and improve function in sensorimotor areas of the brain (de Sousa Fernandes 2020; Wun 2021).

Alternatively to these models based on biomedical concepts, one can argue that exercises could modify some risk factors for developing LBP, such as poor general health, obesity, sedentary lifestyle, occupational exposures, stress, and psychosocial predictors (Dario 2015; Parreira 2018; Taylor 2014). One systematic review found that factors such as chronic diseases, sleep disorders, pain in different body regions, and feeling tired are risk factors for LBP (Taylor 2014). Consequently, exercise programmes could promote positive effects on these factors and support behaviour changes indirectly. Moreover, exercises can affect variables such as fear, catastrophising, self‐efficacy, beliefs, improved psychological well‐being, and cognitive function (Booth 2017; Wun 2021).

Why it is important to do this review

The prevention of LBP could have a direct impact on health and can help reduce the burden of long‐term disability in the general population (Bell 2009; Ferreira 2021a). A comprehensive evaluation of the effects of exercise alone or combined with education is necessary to inform clinical practice and policy decisions. Exercise and education are commonly recommended as prevention strategies, yet there is currently limited high‐quality evidence regarding their effectiveness. New studies have been conducted over recent years, which might improve the certainty of evidence. Additionally, the previous systematic reviews have limitations, such as the inclusion of studies with symptomatic participants at the entry of the study (Bigos 2009; Choi 2010), and the lack of assessment of the certainty of the evidence to interpret the findings (e.g. using GRADE) (Huang 2020). This Cochrane review will provide a rigorous and up‐to‐date synthesis of the available evidence to guide the development of effective prevention strategies for non‐specific LBP.

Objectives

To evaluate the effects of exercise alone or exercise plus education compared with inactive control or education alone to prevent non‐specific LBP.

Methods

Criteria for considering studies for this review

Types of studies

We will include randomised controlled trials (RCTs) and cluster‐RCTs.

Types of participants

We will include studies with adults of both sexes, aged 18 years or greater, who meet one or more of the following criteria:

• healthy participants with no history of LBP;

• participants without LBP when entering the study or neither actively seeking treatment nor disabled from work because of LBP at baseline;

• participants who were recovered from an episode of LBP, with no pain for at least one month; and

• participants with mild LBP, defined as LBP of less than 3 points on a 0 to 10 Numerical Rating Scale (NRS).

We will include studies that recruited participants in different settings, such as hospitals, health centres, outpatient, community, and occupational settings.

LBP is defined as pain or discomfort in the region between the lower edge of the rib 12 and the gluteal fold, generating limitations in usual activities or changes in daily routine for more than one day, and may include pain referring down the lower limbs (Dionne 2008).

We will exclude studies with pregnant populations, participants with severe symptoms of LBP, participants with specific LBP disorders (i.e. trauma, infection, metastatic diseases, or rheumatoid arthritis) or specific spine pathologies (such as radicular pain, radiculopathy, or spinal stenosis). We will exclude studies in which LBP is mixed with other musculoskeletal conditions and where there are no separate data for only those with LBP.

Types of interventions

We will include intervention studies examining exercise alone or associated with education for the prevention of LBP. Exercises are defined as a subset of physical activity that is planned, structured, and repetitive, and has the objective of improving or maintaining one or more components of physical performance (i.e. health‐ or skill‐related). This can be performed with or without the supervision of a health professional or other. We will include all types of exercises, which can be categorised into:

• aerobic exercises (exercises that aim to increase the capacity of the circulatory and respiratory systems to supply oxygen for physical activities, such as walking, cycling, running, etc.);

• resistance exercises (exercises that improve muscle capacity to resist fatigue in sustained physical activity);

• strength exercises (encompasses all exercises that aim to increase muscle capacity to produce force, such as weight training, dumbbell or shin training, elastic band training, core strengthening, etc.);

• flexibility exercises (set of exercises that improve joint mobility, such as static or dynamic stretch);

• neuromotor exercises (exercises that aim to improve agility, balance, or co‐ordination); and

• mixed or multimodal programmes (that involve more than one type of exercise) (ACSM 2018; WHO 2020).

RCTs investigating exercises combined with other interventions will be included if the effect of exercise could be isolated (e.g. exercise and manual therapy versus manual therapy).

Patient education will be broadly defined as any set of planned educational activities aimed at facilitating the person's knowledge base, thus improving the person's health behaviour or health status (or both) (Louw 2018). We will not include education that forms part of a broader psychological intervention (e.g. cognitive behavioural therapy, acceptance and commitment therapy).

We will include trials where exercise alone or combined with education is compared to:

• education alone; or

• exercise alone; or

• inactive control (i.e. placebo, no treatment, usual care, or a non‐exercise minimal intervention, e.g. booklet).

Types of outcome measures

We will not exclude studies based on the presence or absence of outcomes of interest, ensuring a comprehensive assessment of the available evidence.

Primary outcomes

• Disability as a continuous outcome (in the following order of preference: Roland‐Morris Disability Questionnaire (RMDQ), the Oswestry Disability Index (ODI), Patient Specific Functional Scale (PSFS), other validated scale).

• Pain intensity as a continuous outcome (in the following order of preference: Numerical Rating Scale (NRS), Visual Analogue Scale (VAS), McGill Pain Questionnaire, other validated scale).

• New episode of LBP as a dichotomous measure.

Secondary outcomes

• Pain as a dichotomous outcome.

• Disability as a dichotomous outcome.

• Care seeking for LBP (e.g. primary care consultations, physiotherapists), treated as dichotomous data.

• Sick leave treated as a dichotomous measure.

• Health‐related quality of life in the following order of preference: 12‐Item Short Form Health Survey (SF‐12), European Quality of Life Survey‐5 Dimensions (EQ‐5D), the 36‐Item Short Form Health Survey (SF‐36), WHOQOL‐BREF (World Health Organization Quality of Life Scale), PROMIS‐GH‐10 (Patient‐Reported Outcomes Measurement Information System Global Health), and other algofunctional scale. For the SF‐36, SF‐12, and PROMIS‐GH‐10, we will separately report the physical and mental component scores. It will be treated as a continuous variable.

• Adverse events (number and nature of any adverse events, including withdraws and serious adverse events) treated as a dichotomous variable.

We will categorise the follow‐up time for measuring outcomes as short‐term (12 months or less) and long‐term (more than 12 months). If studies use multiple time points, we will prioritise those closer to 12 months for short term, and those closer to three years for long term. Given that non‐specific LBP is characterised by fluctuations in symptoms and potential recurrences over time, a period of 12 months is deemed reasonable to capture the initial effects of interventions and their sustainability. This time frame allows us to comprehensively evaluate the impact of prevention strategies over a period that is sufficiently long to observe meaningful changes in the condition's trajectory, reflecting both short and sustained effects of interventions.

Search methods for identification of studies

Electronic searches

We will search the following databases from inception to current for relevant studies.

• Cochrane Central Register of Controlled Trials (via OvidSP, from inception to current issue)

• MEDLINE (via OvidSP, 1946 to current)

• Embase (via OvidSP, 1980 to current)

• Cumulative Index to Nursing and Allied Health Literature (CINAHL, via EBSCO, 1982 to current)

• ClinicalTrials.gov (clinicaltrials.gov)

• World Health Organization International Clinical Trials Registry Platform (WHO ICTRP; trialsearch.who.int)

We will use the search strategies developed by the Cochrane Back and Neck Review Group. We will include all languages and inclusion criteria in the search strategies. We will use the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE, Embase, and CINAHL (Higgins 2022).

A draft search strategy for MEDLINE can be found in Appendix 1. The strategy will be adapted as closely as possible across the other databases.

Searching other resources

We will screen the reference lists of relevant systematic reviews for additional studies. We will search for errata or retractions from included studies published in full text on PubMed and report the date this was done. We will not search grey literature.

Data collection and analysis

Selection of studies

One review author will run the electronic searches and two review authors will independently screen the titles and abstracts against inclusion and exclusion criteria. We will obtain the full text of articles selected in this phase and articles for which there is disagreement so that the final decision will be based on the full paper. We will resolve disagreements by consensus concerning the final inclusion of RCTs after full‐text review and consult a third review author if disagreements persist. We will list the articles that are excluded at the full‐text stage in the Characteristics of excluded studies table.

Data extraction and management

Two review authors will independently extract the data using a standardised extraction form and resolve all discrepancies by discussion and mediated by a third review author. The data extraction will be based on the study and participants' characteristics, settings, description of interventions/control groups (type of exercises, frequency and duration, specific aspects of the education programme), follow‐up periods, and results of outcomes of interest (number of events and number of participants per treatment group for dichotomous outcomes, and means, standard deviations, and number of participants per treatment group for continuous outcomes). We will collect the reporting of interventions according to the Template for Intervention Description and Replication (TIDieR) checklist (Hoffman 2014; Yamato 2016). We will also collect information regarding funding, and notable declarations of interest of trial authors. We will give preference to intention‐to‐treat analysis data rather than per‐protocol or as‐treated, if available. We will prioritise data from change scores (from baseline) if both change and endpoint values are available. We will use Covidence software for the screening and data extraction phases (Covidence 2024; Higgins 2022; Li 2015).

Assessment of risk of bias in included studies

Two review authors will independently conduct the risk of bias assessment. We will arrive at a consensus if disagreements occur. If disagreement persists, a third review author will arbitrate. To assess the risk of bias, we will remove the name of authors, institutions, and journals, and mask all the studies for results and conclusions. We will conduct risk of bias assessments using the guidelines recommended by the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022), the Cochrane RoB 2 tool for RCTs, the variant 'RoB 2 tool for cluster‐randomised trials', and an Excel tool to implement RoB 2.

When considering treatment effects, we will take into account the risk of bias for the studies that contribute to that outcome. We will assess the risk of bias for the effect of assignment to the intervention (e.g. the intention‐to‐treat effect) for the primary outcomes (disability, pain intensity, and new episode of LBP) for the main comparisons (education and inactive control) at the short‐term (12 months or less). The types of biases and domains assessed will be the following.

• Bias arising from the randomisation process. This domain addresses the following issues: the allocation sequence was random, the allocation sequence was adequately concealed, and the baseline differences between intervention groups suggest a problem with the randomisation process.

• Bias due to deviations from intended interventions. This topic includes if participants were aware of their assigned intervention during the trial and if carers and people delivering the interventions were aware of participants' assigned intervention during the trial.

• Bias due to missing outcome data. This type of bias addresses the following issues: data for this outcome were available for all, or nearly all, participants randomised, there was evidence that the result was not biased by missing outcome data, and missingness in the outcome was likely to depend on its true value.

• Bias in the measurement of the outcome. Here, the issues addressed are: the method of measuring the outcome was appropriate, measurement of the outcome could have differed between intervention groups, outcome assessors were aware of the intervention received by study participants, and assessment of the outcome was likely to have been influenced by knowledge of intervention received.

• Bias in the selection of the reported result. The issues assessed in this type of bias are: the trial was analysed in accordance with a prespecified plan that was finalised before unblinded outcome data were available for analysis, the numerical result being assessed was likely to have been selected from multiple outcome measurements, and the numerical result being assessed was likely to have been selected from multiple analyses of the data (Higgins 2022; Sterne 2019).

For each domain, we will use the resources provided by the RoB 2 tool, which include a series of signalling questions, an algorithm that maps responses to the signalling questions to a proposed judgement and free‐text boxes to justify the responses to the signalling questions. Each of these criteria will be categorised as 'Low risk of bias', 'Some concerns', or 'High risk of bias', and then entered into the risk of bias table (Higgins 2022; Sterne 2019) (see Appendix 2). We will judge an 'overall risk of bias' for each outcome of the trial (see Appendix 3), and use it to feed into the GRADE risk of bias domain assessment.

Measures of treatment effect

We will analyse dichotomous data as risk ratios (RR) or Peto odds ratios when the outcome is a rare event (approximately less than 10%), and provide 95% confidence intervals (CIs). We will analyse continuous data preferably as mean differences (MD) and 95% CIs. For the outcomes of pain and disability, we will convert the scales used to a common 0 to 100 scale to maintain results expressed as MDs and 95% CIs. When studies use different scales to measure the same conceptual outcome (e.g. health‐related quality of life), we will report standardised mean differences (SMD), back‐translating them to a typical scale (e.g. 0 to 10 for pain) by multiplying the SMD by a typical among‐person standard deviation (e.g. the standard deviation of the control group at baseline from the most representative trial). We will enter data presented as a scale with a consistent direction of effect across studies. We will assume a clinically important effect with a mean change score of less than 10 points as very small, 10 to 19 points as small, 20 to 29 points as moderate and greater than 30 points as large on a 100‐point scale for pain and disability (Abdel 2021; Abdel 2023).

For studies reporting time‐to‐event data, such as time to new occurrence of LBP, we will also employ survival analysis methods to summarise these data appropriately and express the intervention effect as a hazard ratio (HR) with 95% CIs. This approach considers both the timing of events and the censoring of data, where participants who do not experience the event within the study period are still accounted for in the analysis. The HR will allow us to compare the instantaneous risk of events between intervention and control groups over the follow‐up period.

In the Effects of interventions results section and the Comments column of the Summary of findings table, we will report the absolute percentage difference, the relative percentage change from baseline, and for outcomes that show a clinically important difference between treatment groups, we will report the number needed to treat for an additional beneficial outcome (NNTB), or number needed to treat for an additional harmful outcome (NNTH). For dichotomous outcomes, we will calculate the NNTB or NNTH from the control group event rate and the relative risk using the Visual Rx NNT calculator (Cates 2008). We will calculate the NNTB for continuous measures (when clinically important) using the Wells calculator (available at the Cochrane Musculoskeletal Group Editorial office, musculoskeletal.cochrane.org).

Unit of analysis issues

The participant will be considered the unit of analysis for all trials. Where multiple trial arms are reported in a single trial, we will include only the relevant arms. If two comparisons are combined in the same meta‐analysis, we will halve the control group to avoid double‐counting (e.g. two groups of different telerehabilitation exercise interventions and one waiting‐list control). If we identify cluster‐RCTs, we will multiply the standard error of the effect estimate (from an analysis ignoring clustering) by the square root of the design effect (inflated variances), according to the method described in Chapter 23 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022). The meta‐analysis using the inflated variances will be performed using the generic inverse‐variance method.

Dealing with missing data

We plan to adopt the following strategies, based on the methods outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022).

• We will contact the authors of studies with incomplete data to request information on missing data. If we receive no reply within six weeks, we will deal with missing data using the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022).

• We will perform sensitivity analyses to assess how sensitive results are to reasonable changes in the assumptions that are made.

• We will address the potential impact of missing data on the findings of the review in the Discussion section.

Assessment of heterogeneity

We will consider clinical diversity, methodological diversity, and statistical heterogeneity to inform the decision of whether to conduct a meta‐analysis. We will assess statistical heterogeneity by visual inspection of the forest plot to assess for clear differences in results between the studies and using the I² and Chi² statistics.

Thresholds for the interpretation of the I² statistic can be misleading since the importance of inconsistency depends on several factors (Higgins 2022). An approximate guide to interpretation will be as follows:

• 0% to 40%: might not be important;

• 30% to 60%: may represent moderate heterogeneity;

• 50% to 90%: may represent substantial heterogeneity;

• 75% to 100%: may represent considerable heterogeneity.

We will interpret the Chi² test so that a P value of 0.10 or less indicates evidence of statistical heterogeneity.

Assessment of reporting biases

We will use funnel plots if we include 10 or more studies in a comparison to identify potential small‐study effects, which may be a marker of publication bias. To assess outcome reporting bias, we will check trial protocols against published reports. For studies published after 1 July 2005, we will screen the clinical trial register at the International Clinical Trials Registry Platform of the World Health Organization (trialsearch.who.int) for the a priori trial protocol. We will also evaluate whether selective reporting of outcomes is present.

Data synthesis

We will use a random‐effects model for all meta‐analyses only where meta‐analysis is meaningful (i.e. if the treatments, participants, and the underlying clinical question are similar enough for pooling to make sense). If meta‐analysis is not possible, we will use alternative synthesis methods as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022).

Subgroup analysis and investigation of heterogeneity

If there are sufficient data, we will conduct the following subgroup analyses.

• Type of population (group 1: healthy participants with no history of LBP; group 2: participants without LBP when entering the study or at least one outcome not present at baseline and participants who were recovered from an episode of LBP, with no pain for at least one month; group 3: participants with mild LBP, defined as LBP less than 3 points in a 0 to 10 NRS).

• Type of exercises (cardiorespiratory/aerobic exercises, resistance exercises, flexibility exercises, neuromotor exercises, and mixed exercises programmes).

We will use the formal test for subgroup interactions in Review Manager (RevMan 2024), and will use caution in the interpretation of subgroup analyses as advised in the section of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022).

Sensitivity analysis

The primary analysis will include all eligible studies. We will re‐analyse the results including only studies at low risk of bias and some concerns to determine if risk of bias altered the synthesised results.

Summary of findings and assessment of the certainty of the evidence

We will create a summary of findings table that will contain: PICO (population, intervention(s), comparison(s) and outcomes), a list of main outcomes, illustrative comparative risks, the magnitude of the effect (relative and absolute effects), the number of participants and studies, the grade of the certainty of the evidence, comments, and footnotes. The table will include the following outcomes.

• Disability

• Pain intensity

• New episode of LBP

• Adverse events

The main comparisons will be exercise intervention alone versus inactive control and exercise intervention plus education versus education plus inactive control (see Table 1).

1. Example of summary of findings table.

Exercises alone versus inactive control for prevention of non‐specific low back pain in mixed population
Participants or population: 'mixed' population Settings: all settings Intervention: exercises alone Comparison: inactive control
Outcomes	Outcome type (continuous/dichotomous)	Outcome measure	Comments (short‐term/ long‐term)
Disability	Continuous	Mixed (RMDQ and ODI)	Short‐term
Pain intensity	Continuous	VAS	Short‐term
New episode of low back pain	Dichotomous	Yes/no	Short‐term
Adverse events	Dichotomous	Yes/no	Short‐term
ODI: Oswestry Disability Index; RMDQ: Roland‐Morris Disability Questionnaire; VAS: Visual Analogue Scale.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

Regardless of whether there are sufficient data available to use quantitative pooling to summarise the data, we will assess the overall certainty of the evidence for each outcome using the GRADE approach, as recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022). Two review authors will independently assess the certainty of the evidence and a third review author will resolve disagreements. The certainty will be based on five domains: limitations in the design and implementation (risk of bias), inconsistency (heterogeneity), indirectness (inability to generalise), imprecision (insufficient or imprecise data), and publication bias. We will downgrade the certainty of the evidence for a specific outcome according to the performance of the studies against these five factors. We will explain our reasons for downgrading or not downgrading in the footnotes of the table (Appendix 4).

Appendices

Appendix 1. MEDLINE search strategy

1. randomized controlled trial.pt.

2. controlled clinical trial.pt.

3. randomi#ed.ab.

4. placebo.ab,ti.

5. drug therapy.fs.

6. randomly.ab,ti.

7. trial.ab,ti.

8. groups.ab,ti.

9. or/1‐8

10. (animals not (humans and animals)).sh.

11. 9 not 10

12. dorsalgia.ti,ab.

13. exp Back Pain/

14. backache.ti,ab.

15. (lumbar adj pain).ti,ab.

16. coccyx.ti,ab.

17. coccydynia.ti,ab.

18. sciatica.ti,ab.

19. sciatica/

20. spondylosis.ti,ab.

21. lumbago.ti,ab.

22. or/12‐21

23. exp Exercise/

24. exercis$.mp.

25. physical fitness.mp.

26. ((strength$ or resist$ or weight$) adj3 training).mp.

27. (yoga or pilates).mp.

28. conditioning.mp.

29. exp Exercise Therapy/

30. exp Exercise Movement Techniques/

31. exp Recreation/

32. exp Physical Fitness/

33. exp Physical Endurance/

34. or/23‐33

35. 11 and 22 and 34

Appendix 2. Risk of bias assessment

We will analyse and judge the risk of bias based on the Cochrane RoB 2 tool (Higgins 2022; Sterne 2019).

1. Risk of bias arising from the randomisation process

1.1 Random sequence generation

Low risk of bias: if sequence generation was achieved using computer random number generator or a random numbers table. Drawing lots, tossing a coin, shuffling cards, and throwing dice are also considered adequate if performed by an independent person.

Some concerns (no information): if the method of randomisation was not specified, but the trial was still presented as being randomised.

High risk of bias: if no random element was used in generating the allocation sequence or the sequence was predictable. Examples include alternation; methods based on dates (of birth or admission); patient record numbers; allocation decisions made by clinicians or participants; allocation based on the availability of the intervention; or any other systematic or haphazard method.

1.2 Allocation concealment

Low risk of bias: if the trial used any form of remote or centrally administered method to allocate interventions to participants, where the process of allocation is controlled by an external unit or organisation, independent of the enrolment personnel, if envelopes or drug containers were used appropriately.

Some concerns (no information): if the trial was classified as randomised but the allocation concealment process was not described.

High risk of bias: if the allocation sequence was familiar to the investigators who assigned participants.

1.3 Baseline differences between intervention groups

Low risk of bias: there were no imbalances apparent or any observed imbalances are compatible with chance.

Some concerns (no information): no useful baseline information available.

High risk of bias: there were imbalances that indicated problems with the randomisation process, such as differences between intervention group sizes, a substantial excess in significant differences in baseline characteristics between intervention groups, and imbalance in one or more key prognostic factors, or baseline measures of outcome variables.

2. Risk of bias due to deviation from intended interventions (performance bias)

This domain refers to the effect of assignment to the interventions at baseline and the effect of adhering to the intervention as specified in the trial protocol.

Low risk of bias: participants, carers, and people delivering the interventions were blinded to intervention groups or, if there was no blinding or incomplete blinding, the review authors judged that the outcome was unlikely to be influenced by lack of blinding and there were no deviations from the intended intervention because of the trial context.

Some concerns (no information): participants, carers, or people delivering the interventions were unblinded and there was no information on whether there were deviations from the intended intervention because of the trial context or there were deviations from intended interventions that arose because of the trial context (but these deviations were unlikely to have affected the outcome or these deviations were balanced between the intervention groups).

High risk of bias: all participants were unblinded to intervention groups during the trial and there were deviations from the intended interventions that arose because of the trial context (but in this case these deviations were likely to have affected the outcome and they were unbalanced between the intervention groups).

3. Risk of bias due to missing outcome data

Low risk of bias: outcome data were available for all, or nearly all, randomised participants or there was evidence that the result was not biased by missing outcome data or missingness in the outcome could not depend on its true value.

Some concerns (no information): outcome data were not available for all, or nearly all, randomised participants, and there was no evidence that the result was not biased by missing outcome data, and missingness in the outcome could have depended on its true value and it is not likely that missingness in the outcome depended on its true value.

High risk of bias: the missingness in the outcome depended on its true value.

4. Risk of bias in measurement of the outcome

Low risk of bias: the method of measuring the outcome was appropriate, and the measurement or ascertainment of the outcome did not differ between intervention groups, and the outcome assessors were blinded or the assessment of the outcome could not have been influenced by knowledge of the intervention received.

Some concerns (no information): the method of measuring the outcome was appropriate, and the measurement or ascertainment of the outcome did not differ between intervention groups or there is no information about that, but the assessment of the outcome could have been influenced by knowledge of the intervention received, but it is unlikely that assessment of the outcome was influenced by knowledge of intervention received.

High risk of bias: the method of measuring the outcome was inappropriate or the measurement or ascertainment of the outcome could have differed between intervention groups or it is likely that assessment of the outcome was influenced by knowledge of the intervention received.

5. Risk of bias in selection of the reported result

Low risk of bias: the data were analysed in accordance with a prespecified plan that was finalised before unblinded outcome data were available for analysis, the result being assessed was unlikely to have been selected, on the basis of the results, from multiple eligible outcome measurements, and reported outcome data are unlikely to have been selected, on the basis of the results, from multiple eligible analyses of the data.

Some concerns (no information): the data were not analysed in accordance with a prespecified plan or there is no information on whether the result being assessed is likely to have been selected, on the basis of the results, from multiple eligible outcome measurements and from multiple eligible analyses of the data.

High risk of bias: the result being assessed is likely to have been selected, on the basis of the results, from multiple eligible outcome measurements or from multiple eligible analyses of the data.

Appendix 3. Overall risk of bias

We will assess overall risk of bias as follows.

• Low risk of bias: if all domains are at low risk of bias.

• Some concerns: if at least one domain is at some concerns, but none are at high risk of bias for any domain.

• High risk of bias: if at least one domain is classified at high risk of bias.

Appendix 4. GRADE approach to evidence synthesis

The certainty of the evidence will be categorised as follows.

• High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.

• Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.

• Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect.

• Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

We will rate the certainty of the evidence for each outcome across the studies according to the factors outlined in the GRADE approach (Schünemann 2013). Considering that we will include only randomised controlled trials and cluster randomised controlled trials, the body of evidence will start with a high‐certainty rating, and we will downgrade according to the following five factors.

• Limitations of detailed design and execution (risk of bias criteria). For randomised trials, the following limitations will result in bias: lack of allocation concealment, lack of blinding, incomplete accounting of participants and outcome events, selective outcome reporting, or other limitations.

• Inconsistency (or heterogeneity). We will explore all possible explanations for heterogeneity and downgrade if a satisfactory explanation cannot be found. It will be downgraded one or two levels depending on the magnitude of the inconsistency of the results.

• Indirectness (PICO (population, intervention(s), comparison(s) and outcomes) and applicability). We will consider differences in population, interventions, comparisons, and outcomes measures.

• Imprecision (number of events and confidence intervals). Usually, results are imprecise when the studies have wide confidence intervals around the estimate of the effect. For dichotomous or continuous outcomes, we will follow the recommendations of the GRADE Handbook based on the optimal information size criterion (Schünemann 2013).

• Publication bias. We will use the following methods to detect the risk of publication bias: visual inspection and tests for asymmetry of funnel plots and the 'trim and fill' method as an extension of the funnel plot. It is challenging to be confident that publication bias is absent or to set a threshold for downgrading the certainty of evidence due to the strong suspicion of publication bias. For this reason, we will downgrade the certainty of the evidence for publication bias by a maximum of one level, as suggested by GRADE (Schünemann 2013).

Contributions of authors

Conception and design: SPSS, BTS, MJH, CGM.

Drafting of the protocol: SPSS, BTS.

Critical revision of the protocol for important intellectual content: BTS, MJH, TSM, CGM.

Final approval of the protocol: all authors.

Sources of support

Internal sources

• None, Other

  The authors received no internal source of support.

External sources

• CAPES, Brazil

  SPSS has received a research grant from Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brasil (CAPES) – Finance Code 001.

• NHMRC Research Fellowship, Australia

  CGM holds a research fellowship from the National Health and Medical Research Council. NHMRC is an Australian government agency that funds health research.

Declarations of interest

SPSS: none.

BTS: none. He is a Cochrane Editor, but was not involved in the editorial process.

CGM: none. He is a Cochrane Editor, but was not involved in the editorial process.

MJH declares the following: grant/contract from National Health and Medical Research Council, payment for lectures and travel costs from Korean Academy Maitland Orthopedic Manipulative Physical Therapy, editorial on prevention of low back pain for BJSM, and Cochrane Back and Neck group associate editorial board. He was not involved in the editorial process.

TSM: none.

New