Objective To evaluate the evidence from randomised controlled trials (RCTs) on the effectiveness of prevention strategies to reduce future impact of low back pain (LBP), where impact is measured by LBP intensity and associated disability.
Design Systematic review with meta-analysis.
Data sources MEDLINE, Embase, CINAHL, PEDro and The Cochrane (CENTRAL) databases from inception to 22 October 2018.
Eligibility criteria RCTs evaluating any intervention aiming to prevent future impact of LBP, reporting an outcome measure of LBP intensity and/or disability measured at least 3 months post-randomisation, and the intervention group must be compared with a group that received no intervention/placebo or minimal intervention. Trials restricting recruitment to participants with current LBP were excluded.
Results 27 published reports of 25 different trials including a total of 8341 participants fulfilled the inclusion criteria. The pooled results, from three RCTs (612 participants), found moderate-quality evidence that an exercise programme can prevent future LBP intensity (mean difference (MD) −4.50; 95% CI −7.26 to −1.74), and from 4 RCTs (471 participants) that an exercise and education programme can prevent future disability due to LBP (MD −6.28; 95% CI −9.51 to −3.06). It is uncertain whether prevention programmes improve future quality of life (QoL) and workability due to the overall low-quality and very low-quality available evidence.
Conclusions This review provides moderate-quality evidence that an exercise programme, and a programme combining exercise and education, are effective to reduce future LBP intensity and associated disability. It is uncertain whether prevention programmes can improve future QoL and workability. Further high-quality RCTs evaluating prevention programmes aiming to reduce future impact of LBP are needed.
- lower back
- randomised controlled trial
Statistics from Altmetric.com
Low back pain (LBP) is the leading cause of global disability and a common reason for work absenteeism, lost productivity and care-seeking.1–3 Although most people with an episode of LBP improve substantially within 6–12 weeks,4 most will also experience a recurrence within 12 months.5 For this reason, LBP is considered to be a chronic condition with recurrent symptomatic episodes. Effective prevention strategies to reduce future LBP intensity and associated disability have the potential to greatly reduce the burden associated with this condition.
A recent systematic review6 reported moderate-quality evidence that exercise combined with education reduces the risk of a future episode of LBP (relative risk 0.55; 95% CI 0.41 to 0.74), but that most other interventions either lacked evidence or appeared to be ineffective. Importantly, this review took a traditional approach to prevention by only including studies recruiting participants without LBP at baseline. While this approach works well in acute conditions where the onset and the end of the episode are clear, it has limitations for a chronic recurrent condition like LBP. In chronic fluctuating conditions, it is arguably more important to prevent the consequences of the chronic disease (sometimes considered tertiary prevention) than to simply prevent future episodes.
The effectiveness of prevention strategies in terms of reducing future LBP intensity and/or associated disability rather than preventing future episodes of LBP have been explored in previous studies.7 8 Studies such as these commonly include ‘mixed populations’ (ie, both asymptomatic and symptomatic patients) at study entry, rather than restricting inclusion to people without LBP. These studies provide important information about the potential effectiveness of prevention strategies on reducing future LBP intensity and associated disability, but were not included in the previous systematic review.6 Studies including ‘mixed populations’ also differ from traditional treatment studies that require all participants to have symptoms at study entry. It is timely to conduct a synthesis of randomised controlled trials (RCTs) to enable better understanding of the effects of prevention strategies to reduce future impact of LBP.
Therefore, the primary aim of this systematic review was to investigate the effectiveness of prevention strategies aiming to reduce future impact of LBP, where impact is measured by LBP intensity and associated disability.
Study reporting and protocol registration
The systematic review adhered to the statement for reporting systematic reviews and meta-analyses of studies that evaluate healthcare interventions (Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)).9 The review protocol was prospectively registered on PROSPERO (CRD42018107946).
Data sources and searches
A comprehensive search of five electronic databases (MEDLINE via Ovid, EMBASE via Ovid, CINAHL, Physiotherapy Evidence Database (PEDro) and The Cochrane Central Register of Controlled Trials (CENTRAL) via The Cochrane Library for eligible manuscripts was conducted from the date of inception to 22 October 2018. A sensitive search strategy was used based on the recommendations of the Cochrane Back and Neck Group10 for ‘randomised controlled trials’ and ‘low back pain’, combined with search terms for ‘prevention’. The full search strategy for each database is presented in the online supplementary appendix A. In addition, reference lists of relevant reviews and included RCTs were manually searched and citation tracking of all included trials was performed. The searches and inclusion criteria were not restricted by language.
Study selection and screening criteria
We included published reports of RCTs, including cluster-RCTs, testing the effectiveness of prevention strategies aiming to reduce future impact of LBP. Impact of LBP was measured by LBP intensity and disability. We excluded RCTs that restricted recruitment to participants with current LBP (treatment studies). Eligible interventions included any approach aiming to prevent or reduce future impact of LBP such as workplace interventions to control risk factors or interventions to make people more fit/healthy/resilient. To be eligible, trials needed to compare an intervention group with a group that received no intervention, sham intervention or minimal intervention. We also included RCTs investigating multimodal interventions if the effect of one intervention could be isolated (eg, back exercise and education vs education alone).
Trials needed to report an outcome measure of LBP intensity and/or LBP-associated disability measured at least 3 months postrandomisation. Primary outcomes for this review were: a) pain-intensity measured by a self-reported outcome measure (eg, visual analogue scale, numerical rating scale) and b) disability measured by a self-reported outcome measure (eg, Oswestry Disability Index and Roland-Morris Disability Questionnaire). Other patient-centred outcomes relevant to back pain such as quality of life (QoL) were considered secondary outcomes. Studies that used a quasi-randomised design were excluded.
A three-stage screening process was used to select relevant RCTs for this review. In the first stage, one reviewer (TFdC) screened all titles for eligibility and excluded clearly irrelevant studies. In the second stage, each study title and abstract was independently evaluated by pairs of review authors (TFdC, DS, JTF, MH, SA). In the third stage, the full-text for each potentially eligible study was assessed against the eligibility criteria by a pair of independent review authors (TFdC, DS, JTF, MH, SA). Disagreements were resolved through discussion. We contacted authors for additional information as necessary.
Data for each included trial were extracted by pairs of independent reviewers (TFdC, DS, JTF, MH, SA) using a standardised data extraction form and discrepancies resolved through discussion. Extracted data included: study characteristics (eg, source, study design, country, participant’s characteristics, outcome measure, description of the intervention/control groups and follow-up periods), means and measures of variability for all outcomes. When possible, raw mean and SD outcome data for both the intervention group and control group were extracted. We also estimated raw data from graphs in cases where this information was not presented in tables or text. We attempted to contact authors of included RCTs to clarify any relevant information or request additional data when required.
Risk of bias was assessed using the PEDro scale11–13 by either downloading the available scores from the PEDro database (http://www.pedro.org.au), or by two experienced PEDro raters rating the report when not available online. The total score on the PEDro scale is the addition of ‘yes’ (criterion is clearly satisfied) responses for items 2–11 (item 1 is not used for calculation of the total PEDro scale score because it is more related to external validity) and range from 0 (high risk of bias) to 10 (low risk of bias). The PEDro scale total score has acceptably high reliability and validity11 12 and Rasch analysis has confirmed that it can be used as a continuous scale.14
Quality of evidence assessment
The overall quality of evidence for each intervention contrast was rated as high, moderate, low or very low quality as recommended by the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system.15 The GRADE classification was downgraded one level per study limitation, from high quality, if any of the following limitations were present: (i) design limitation (more than a quarter of participants from studies with low methodological quality (PEDro score <7)); (ii) inconsistency of results (wide variation of point estimates across individual trials or substantial heterogeneity, I2 >50%); (iii) imprecision (based on a threshold of <400 total participants for each pooled outcome estimate). We did not consider the indirectness criterion in this review as we included a specific population with relevant outcomes. When only a single RCT was available, evidence from RCTs with fewer than 400 participants was downgraded for inconsistency and imprecision; however, evidence from single RCTs presenting >400 participants was only downgraded for inconsistency. Publication bias was not evaluated due to the small number of trials in each meta-analysis.16
A GRADE profile was completed for each pooled estimate and for single RCTs comparing a LBP prevention strategy with a control intervention. Two independent reviewers (TFdC and MH) independently performed GRADE assessments for each treatment contrast and disagreements were resolved by discussion.
The between-groups mean difference (MD) and 95% CIs were calculated using the mean final score for the intervention and control groups. We used final scores rather than within group change scores as only one study reported change scores.16 When possible, we combined results in a meta-analysis using random-effects models. Negative values of the mean difference estimate represent an effect in favour of the intervention group. To accommodate the different scales used for study outcomes, we converted, whenever possible, outcomes to a common 0–100 scale. If conversion was not possible due to the nature of outcome (eg, categorical or ordinal), we did not convert the results but instead presented them as a narrative synthesis. If information regarding SD was missing, we calculated these from CIs, SEs or p values; however, if no measure of variability was presented, we estimated the SD from the most similar and high-quality trial in the review as recommended by The Cochrane Collaboration.16
Outcome assessment data were extracted for two time periods: short-term follow-up (collected at <12 months postrandomisation); long-term follow-up (collected at ≥12 months postrandomisation). When studies presented multiple follow-up time-points that fell within the same category, we used the time-point that was closest to 6 months for short-term follow-up and one closest to 12 months for long-term follow-up. For RCTs including multiple treatment arms, we extracted data for each comparison that met the inclusion criteria and adjusted the numbers per group as recommended by The Cochrane Handbook for Systematic Review of Interventions.16
Trials considered homogeneous were grouped, when possible, according to the population (eg, children, pregnant women), intervention strategy, outcome measure and outcome assessment time-points (short-term and long-term). For RCTs not reporting the sample size at the follow-up time-point, we adopted the baseline sample size.
Where we considered study interventions to be sufficiently similar to be combined in meta-analyses, we assessed heterogeneity of treatment effects by visual inspection of effect size with 95% CI and by using the I2 statistic. We used Comprehensive Meta-Analysis, V.2.2.064 (Biostat) for all analyses.
Of the 17 342 identified records, 176 were considered potentially eligible and we reviewed full-text manuscripts. Of these, 27 published reports (25 different RCTs including 8341 participants) met the inclusion criteria and were deemed eligible for this review.7 8 17–41 The 25 RCTs included 10 cluster-RCTs.7 17 20 28 30 34 35 39–41 Two RCTs were reported in four published manuscripts, with two manuscripts32 36 reporting on short-term follow-up and the other two manuscripts25 37 reporting on long-term follow-up.25 32 36 37 An outline of the screening and selection process is provided in figure 1.
The included studies investigated three different populations: general adults, pregnant women and children. Most trials recruited participants who were employees at a hospital (32%) or company (40%) setting while only two trials (8%) recruited people from the general community. Most included trials (8269 participants) examined a working-age population with the mean age of 45.1 years and majority female (75.9%). Six different LBP prevention strategies were investigated: exercise, exercise and education, education, ergonomics, ergonomics and education and lumbar support. Two trials investigated LBP prevention strategies in a population of pregnant women,21 23 while one trial investigated a sample of primary school children.22 Eight trials presented two intervention contrasts (three arms).7 17 19 29 31 35 40 41 Table 1 and online supplementary appendix B provide details of the characteristics of each included trial.
Risk of bias scores for 247 8 17–28 30–32 34–37 39–41 of the included studies were found on the PEDro database website. The other three studies29 33 38 were assessed by two raters. The mean (SD) PEDro score was 5.4 (1.2) with blinding, concealed allocation, intention-to-treat analysis and adequate follow-up being the main items scored as high risk of bias in 92%, 63%, 55% and 52% of included studies, respectively. The PEDro scale ratings for individual items and the total score for each included RCT are available in online supplementary appendix C.
Raw final scores data for intervention and control groups were available for 23 of the 25 included trials. For the remaining two trials, we used the reported MD (95% CI).17 28 For six trials21 31 34–36 39 we calculated SD and for two trials20 38 we imputed data from similar studies. Study design, follow-up time-point, outcome measure, sample size, raw MD and SD for each intervention and between-groups MD (95% CIs) for all included trials are presented in online supplementary appendix D (primary outcomes) and online supplementary appendix E (secondary outcomes). Trials were grouped according to the prevention strategy, outcomes, follow-up time-point (short-term or long-term) and population. Table 2, online supplementary appendix F (primary outcomes), online supplementary appendix G and online supplementary appendix H (secondary outcomes) provide a summary of the findings and the quality of evidence (GRADE) rating.
Effectiveness of interventions for primary outcomes
Three trials (612 participants) investigated the short-term effects of exercise programmes on prevention or reduction of future LBP intensity and associated disability.17 27 34 The pooled results of three trials (four intervention contrasts) provided moderate-quality evidence that exercise is effective for preventing future LBP intensity (MD −4.50; 95% CI −7.26 to −1.74) (tables 2 and 3).
For prevention of associated disability due to LBP, a single trial (189 participants) provided very low-quality evidence of no short-term effect of exercise programmes (MD −2.36; 95% CI −7.11 to 2.39) (tables 2 and 4).27
Exercise and education
Three trials (184 participants) investigated the effectiveness of an exercise and education prevention programme on reducing future LBP intensity at short-term follow-up,30 32 36 and four trials8 32 36 40 (471 participants) at long-term follow-up. The pooled results of the three trials provided low-quality evidence that an exercise and education programme is not effective at short-term follow-up (MD −1.95; 95% CI −10.09 to 6.18). The long-term results are based on pooling for the four trials and provided moderate-quality evidence of no long-term effect (MD −4.37; 95% CI −9.16 to 0.43) (tables 2 and 3).
For prevention of future disability due to LBP, two trials (150 participants) investigated short-term follow-up,32 36 and four trials8 32 36 40 (471 participants) long-term follow-up. Pooled results of the two trials provides low-quality evidence of no short-term effect of an exercise and education programme on reducing future disability associated with LBP (MD −4.94; 95% CI −12.78 to 2.90). For long-term follow-up, four trials were pooled and provided moderate-quality evidence that exercise and education programme is effective to reduce future disability associated with LBP (MD −6.28; 95% CI −9.51 to −3.06) (tables 2 and 4).
The short-term effect of an education programme on preventing future LBP intensity was investigated in four trials,18 29 31 33 while two trials7 40 reported results on long-term effects. The pooled results of three trials (777 participants)18 29 33 provided moderate-quality evidence that education programmes do not prevent future LBP intensity at short-term follow-up (MD −1.81; 95% CI −4.68 to 1.07). One trial (57 participants)31 was not included in the meta-analysis as it was not possible to convert data to a 0–100 scale. The long-term results are based on pooling of the two trials (126 participants)7 40 and provide low-quality evidence of no effect (MD 1.71; 95% CI −6.14 to 9.56) (tables 2 and 3).
For prevention of LBP-associated disability, four trials (804 participants)29 33 38 41 reported short-term data, and two trials (176 participants)40 41 reported long-term data. The pooled results of the four trials provide moderate-quality evidence of no short-term effect (MD −2.59; 95% CI −6.15 to 0.96), while pooling of the two trials provide low-quality evidence of no long-term effect (MD −0.29; 95% CI −4.87 to 4.30) (tables 2 and 4).
Three trials20 24 31 investigated the effectiveness of an ergonomics programme on prevention of future LBP intensity at short-term follow-up (619 participants), and a single trial20 at long-term follow-up (538 participants). It was not possible to pool estimates for the three trials investigating short-term follow-up as we could not convert two trials24 31 to a 0–100 scale. The results from one trial20 on short-term (552 participants) (MD 1.40; 95% CI −3.28 to 6.08) and long-term (538 participants) (MD 2.00; 95% CI −2.74 to 6.74) follow-ups provide low-quality evidence of no effect on preventing future LBP intensity (tables 2 and 3).
Ergonomics intervention and education
The effectiveness of an ergonomics and education programme for preventing future LBP intensity (short-term) and LBP-associated disability (short-term and long-term) was investigated in a single trial.28 The results from one trial on short-term (192 participants) effect for either prevention of future LBP intensity (MD 1.00; 95% CI −6.93 to 8.93) or disability due to LBP (MD 2.08; 95% CI −1.87 to 6.03), and long-term (184 participants) effect for disability due to LBP (MD 1.25; 95% CI −3.08 to 5.58) provide very low-quality evidence of no preventive effect.28 The long-term effect on preventing future LBP intensity was investigated in two trials (266 participants)7 28 and provide low-quality evidence of no effect (MD 0.00; 95% CI −6.70 to 6.70) (tables 2–4).
Effectiveness of interventions for primary outcomes in special populations
Three trials investigated the short-term effect of two different strategies to prevent future LBP intensity and associated disability in pregnant women and children.21–23 Pooling of two trials (452 participants) provides moderate-quality evidence that an exercise programme was not effective for prevention of future LBP intensity (MD −2.70; 95% CI −6.56 to 1.17) at short-term follow-up in pregnant women.21 23 In addition, one trial (240 participants) provides low-quality evidence of no preventive effect on future disability due to LBP (MD −2.91; 95% CI −7.06 to 1.24) in pregnant women.21
Furthermore, a single trial (70 participants) shows very low-quality evidence that an exercise and education programme has no effect on preventing future LBP intensity (MD 0.00; 95% CI −11.68 to 11.68) in children at short-term follow-up.22 Results are presented in tables 2–4.
Effectiveness of interventions for secondary outcomes
Four secondary outcome measures (QoL, workability, pain duration and duration of sick leave) were investigated in 18 included trials8 17 19–22 24–30 32 35–39 41; however, only two outcomes (QoL and workability) were included in the meta-analysis as we could convert data to a 0–100 scale. Overall, we found the evidence was low quality or very low quality with intervention contrasts suggesting no prevention effect on either QoL or workability at short-term and long-term follow-ups. Results for secondary outcomes are presented in online supplementary appendix G and online supplementary appendix I.
Moderate-quality evidence based on three trials (612 participants) (4 intervention contrasts) indicates that exercise alone can reduce future LBP intensity (MD −4.50; 95% CI −7.26 to −1.74) at short-term follow-up. We found no studies that investigated the long-term prevention effect of exercise. Moderate-quality evidence from four trials (471 participants) indicates that exercise and education programmes can reduce future disability associated with LBP (MD −6.28; 95% CI −9.51 to −3.06) at long-term follow-up. In addition, although not statistically significant, the evidence for exercise and education suggests that at short-term it may reduce future disability associated with LBP (MD −4.94; 95% CI −12.78 to 2.90), and at long-term it may reduce future LBP intensity (MD −4.37; 95% CI −9.16 to 0.43). It is uncertain whether education, ergonomics and ergonomics combined with education or interventions delivered in special populations (ie, pregnant women and children) can reduce future LBP intensity and associated disability due to very low quality to low quality of evidence. The impact of prevention programmes on the outcomes of QoL and workability is also unclear due to very low-quality to low-quality evidence.
Previous research demonstrates that exercise programmes alone or in combination with education have moderate-quality evidence supporting their effectiveness to reduce the risk of a future episode of LBP.6 Our review investigated different outcomes (pain intensity and disability, rather than episodes of LBP) and different populations (including some people with current LBP), but our results support the evidence that exercise alone and in combination with education can also reduce future LBP intensity and associated disability.
Our finding that exercise alone and exercise combined with education can reduce future LBP intensity and associated disability respectively was based on moderate-quality evidence, which means further high-quality RCTs are needed. The absolute effect sizes for exercise alone (MD −4.50; 95% CI −7.26 to −1.74) and exercise combined with education (MD −6.28; 95% CI −9.51 to −3.06) appear small; however, these effects must be considered in the context of LBP prevention, where these relative effects across large populations may be important. For instance, when we look at the long-term outcome of disability for exercise combined with education, we found a 20% relative reduction. We suggest that clinicians present to patients the available evidence on strategies to prevent or reduce future LBP, and engage in shared decision making on whether to deliver an exercise and education programme. Factors such as the patient’s enthusiasm and available time to engage in an exercise programme, and their underlying risk of future low back pain should be considered.
Some of the strengths of this study include the use of a prespecified protocol registered on PROSPERO; no inclusion restriction on populations, settings and age; sensitive search strategy using multiple electronic databases with supplementary hand searching, following the PRISMA recommendations; the use of the GRADE system to appraise the overall quality of the evidence and the use of PEDro scale to assess risk of bias of included trials.
The following limitations should be considered when interpreting our results. Despite our best efforts, authors for one potentially eligible RCT42 could not be contacted, and some SDs were not published and had to be estimated from a similar included trial as recommended by The Cochrane Collaboration 16; 9 cluster-RCTs (18 intervention contrasts) required adjustment for clustering; only a small number of trials were included for most intervention contrasts, and some outcome measures (eg, pain duration and duration of sick leave) could not be pooled due to the heterogeneity in measurements. For some trials, the limited descriptions of the experimental intervention and minimal intervention made it difficult to be certain if the control group met our criteria for minimal intervention control. As an example, the control group in the study by Tuchin38 did some exercises; however, these were limited and appeared to be very broad and not specific to spinal pain (‘warm-up stretching programme for sports’). Data inspection suggested that some data were likely skewed (mean/SD<2).43 Therefore, we conducted unplanned sensitivity analyses on the study’s primary outcomes of pain intensity and disability using the log-transformation methods recommended by Higgins et al,44 and included these as online supplementary appendix J and online supplementary appendix K, respectively. Between-group differences on the log-transformed scale were then back-transformed producing effects as ratios with the 95% CI (see online supplementary appendix J and online supplementary appendix K), enabling comparison with the original effects from raw data. The results of these sensitivity analyses were consistent with the original analyses using raw data in terms of effect direction, size and statistical significance, other than the short-term effect on the disability outcome of the intervention contrast comparing education with control, which changed from a small, non-significant, beneficial effect when using original raw data to a small, significant, beneficial effect when using the log-transformed data (see online supplementary appendix K table). Most studies included in our review had sample sizes >50 participants, and therefore inferences based on means are less problematic due to the central limit theorem.44 45
There is moderate-quality evidence indicating that an exercise programme can reduce future LBP intensity at short-term follow-up and that exercise combined with education can reduce future disability due to LBP at long-term follow-up. Interventions including education alone, ergonomics and ergonomics combined with education or interventions for specific populations (ie, pregnant women and children), do not seem to reduce future LBP intensity and associated disability. The impact of prevention programmes on future QoL and workability is unclear due to the low quality to very low quality of available evidence.
What is already known?
The available research suggests exercise combined with education reduces the risk of a future episode of low back pain; however, it is unclear if effective prevention strategies exist to reduce future low back pain intensity and associated disability.
What are the new findings?
We found moderate-quality evidence supporting the effectiveness of exercise as a prevention strategy to reduce future low back pain intensity at short-term follow-up and that exercise combined with education can reduce future disability associated with low back pain at long-term follow-up.
We are uncertain whether prevention strategies can positively impact quality of life or workability owing the low-quality to very low-quality evidence found.
TFdC has a PhD scholarship from Macquarie University (Macquarie University Research Excellence Scholarship (MQRES)). CGM is supported by a National Health and Medical Research Council Principal Research Fellowship.
Twitter @TF_Campos, @CGMMaher, @SOuRCe_RPA
Contributors TFdC and MH had full access to all the data in this systematic review and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: TFdC, CGM, JTF, DS, MH. Acquisition, analysis or interpretation of data: all authors. Drafting of the manuscript: TFdC, CGH, MH. Critical revision of the manuscript for important intellectual content: all authors. Statistical analysis: TFdC, CGM, JTF, MH. Administrative, technical or material support: TFdC, JTF, DS, SA, MH. Study supervision: TFdC, CGH, MH.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.