Studies were deemed eligible for inclusion if they included adult participants with acute pain, using the commonly accepted descriptor for acute pain as being less than 3 months.19,20 Participants with pain greater than 3 months duration or with specific and/or serious spinal pathologies (fracture, spinal haematoma, infection, malignancy, acute cord compression) were excluded. Patients were required to have been stratified according to their level of risk for developing chronic pain or having a poor functional outcome, and then provided a treatment matched to this level of risk. This included all participants from those at low risk of a poor outcome (and hence more likely to have a positive outcome) through to high risk of a poor outcome, provided that they had been stratified. Only randomized-controlled trials (RCTs) were included as systematic reviews of RCTs are considered the highest level of evidence.21,22 For a complete description of the inclusion and exclusion criteria, refer to supplementary material – Table S.1.

A search was conducted across five databases (Medline, Embase, PEDro, CINAHL, and Cochrane Library) from database inception to November 21st, 2022. Studies were limited to humans and the English language. The reference lists of relevant articles were also scanned to identify any eligible studies which may have been missed by the database search. Refer to Table A.1 for an example search strategy. A complete search strategy for each database can be found in the supplementary material.

Two reviewers independently screened titles and abstracts using the web-based Covidence software platform.23 Relevant full texts were then independently reviewed by the two researchers to determine suitability for inclusion. Any conflicts in this process were resolved through consensus discussion between the two researchers with a third researcher available if consensus was unable to be achieved. One researcher independently collected and collated key data from the studies, which was then cross-referenced for accuracy with a second researcher. The extracted data were then converted to table form to allow for easy comparison between studies. Absent data were attempted to be obtained through review of published supplementary resources, search of ClinicalTrials.gov, and through contact with study authors. Authors were given 60 days to respond to written requests for missing information via email. Studies with absent data were assessed on a case-by-case basis. Studies without sufficient data for meta-analysis were analysed descriptively.

Outcome data relating to participant outcomes such as pain and disability were extracted. Outcomes relating to cost and health-care utilization were not extracted as this was beyond the scope of the review. Outcomes were extracted for all available timeframes and then grouped into short-term (< 3 months), intermediate term (3–6 months), and long-term (>6 months). For studies that did not publish raw data relating to the standard deviations of outcomes, these were estimated using calculations outlined in the Cochrane Handbook for Systematic Reviews of Interventions.24 Participant demographic information, settings, method of stratification, and details of interventions, including frequency and duration, were also extracted.

The quality of the included studies was assessed using the PEDro Scale; an 11-item tool that is validated to assess the methodological quality of studies.25 Two reviewers independently appraised articles using this scale, with consensus discussion used for resolution of any conflicts.

Meta-analysis was conducted on trials with sufficient homogeneity, using a random effects model and presented as mean difference. Trials were considered sufficiently homogenous if there was a common outcome measure, comparator, and follow-up timeframe. Review Manager (RevMan) (Version 5.4)26 was used to complete the meta-analysis. Where outcomes of interest (eg. pain, disability) were measured on the same numeric scale (eg. 0 – 100), these were synthesized using a mean difference approach. This was chosen over a standardized mean difference, to enhance comparability of findings. Outcomes for pain were presented using a 0 – 10 scale and outcomes for disability were presented using a 0 – 100 scale. The I2 statistic27 was used to quantify between-study heterogeneity. The results of meta-analysis were compared to pre-determined minimally clinically important differences (MCID), as proposed by Ostelo,28 to establish clinical significance. The MCID was considered to be a change of 2 points of a 0 – 10 scale for pain outcomes and 10 points on a 0 – 100 scale for disability.28

To assess the certainty of evidence, the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach29 was utilised. This process involves evaluating each meta-analysis and downgrading the quality of evidence across five domains. Meta-analyses of RCTs are initially considered high quality and may be downgraded to moderate, low, and very-low. The downgrade criteria were informed by Guyatt 30-34 and are as follows: for risk of bias, analyses were downgraded by one if >25 % of participants were from studies at a high risk of bias (PEDro <6) and by two if >50 % of participants were from studies at high risk of bias. For inconsistency, analyses were downgraded by one if there was moderate heterogeneity (I2 > 40 %) and downgraded by two if there was very substantial heterogeneity (I2 > 75 %). For indirectness, analyses were downgraded by one if there was variation in one of the following, (or downgraded by two if there was variation in more than one): outcome measures used, timeframes, or interventions. For imprecision, analyses were downgraded if there were fewer than 2000 participants included or if the 95 % CI width was greater than twice the minimally clinically important difference; and for publication bias, analyses were downgraded if <10 studies were included. Refer to supplementary material- Table S.2 for a complete summary of the GRADE downgrade criteria utilised.

Our initial search generated 3991 articles with removal of duplicates. From this, 3952 were excluded based on title and abstract screening, leaving 39 which was reduced to 5 following review of full text. Refer to Fig. 1 for the complete PRISMA flowchart. Five studies were included in the analysis, encompassing 3519 participants. Study size ranged from 4735 to 2300 participants36 and were completed in three high-income countries37: United States,36,38,39 Hong Kong SAR (China),35 and Germany.40 Female participants made up, on average, 55 % in each study and only one study36 published demographic data relating to race and ethnicity. Hazard39 did not publish baseline demographic data. Follow-up time ranged from 4 weeks to 12 months, with the majority of studies reporting outcomes between 3 and 6 months. The STarT Back Screening tool was the most common method of risk stratification, used in three studies36,38,40 with the Örebro Musculoskeletal Pain Screening Questionnaire35 and the Vermont Disability Prediction Questionnaire39 used in the remaining two articles. These three tools are validated for predicting risk of poor outcomes; the STarT Back Screening Tool for future disability,15 the Vermont Disability Prediction Questionnaire for return-to-work outcomes,15 and the Örebro Musculoskeletal Pain Screening Questionnaire for both future disability and return-to-work outcomes.15 Priebe et al40 was the only study to also include participants classified as being low risk. Interventions varied across studies. Delitto et al36 and Lee et al35 incorporated treatment aligned with psychological principles, Magel et al38 compared an early intervention with usual care, Priebe et al40 involved consultation with a pain specialist and use of a back pain app and Hazard et al39 involved letters sent to physicians with recommendations. Comparator interventions varied across studies. Hazard et al39 compared risk stratification to no intervention, whereas the remaining studies compared risk stratification to usual care, which was considered management through a primary care clinician on an individual basis. Lee et al35 was the only study to provide specific details on the nature of the usual care provided, incorporating electrophysical, manual, and exercise therapies delivered by a physical therapist. The most common outcomes of interest were pain and disability with only one study including transition to chronic back pain as an outcome measure.36 The full details of the study characteristics are outlined in Table 1.

Fig. 1.

PRISMA flowchart.

(0.26MB).

The study quality ranged from a PEDro score of 3 to 7, with a mean score of 5.4, indicating overall fair quality.41 The risk of bias for each article is outlined in Table 2.

Meta-analysis comparing risk stratification to usual care was completed incorporating 5 cohorts from 4 studies (Fig. 2). Hazard et al39 was the only study to compare risk stratification to no intervention and did not contain sufficient data for quantitative analysis. For these reasons, this study was not included in the meta-analysis.

Fig. 2.

Forest Plot A) Mean Difference: Disability at Intermediate Term (3–6 months) B) Mean Difference: Pain at Intermediate Term (3–6 months).

(0.27MB).

Magel et al,38 Priebe et al,40 and Delitto et al36 shared common outcome measures, taken between 3 and 6 months (intermediate time-frame), and were therefore included in the meta-analysis. For Lee et al,35 the timeframe for assessment of outcomes was not specified, however, participants were assessed at the conclusion of the intervention, which was up to 3 months in duration. Given the proximity to the timeframe of other included studies (3–6 months), this study was therefore included in the meta-analysis. As Magel et al38 included separate outcome data for its medium- and high-risk groups, these were included separately in this meta-analysis despite being part of the same study. Magel et al38 was also the only study to assess outcomes at a short-term (4 weeks) and long-term (12 months) period. As this was limited to one study, no synthesis was possible for these timeframes.

Using the GRADE approach, each meta-analysis was assessed as having very-low certainty. Refer to Table 3 for complete details of the GRADE outcome for each analysis.

Sensitivity analysis was completed by conducting repeat meta-analysis with small methodological alterations. These alterations included removing Priebe et al40 due to its low methodological quality, and repeating meta-analysis using both a random effects model and standardized mean difference. With the removal of Priebe et al40 from the meta-analysis relating to disability, the results demonstrated a statistically significant difference favouring risk stratification (MD= −0.11, 95 % CI: −0.22, −0.00). This contrasts with the original findings. No other sensitivity analysis generated contrasting findings.

This review found very-low certainty evidence that the use of a risk stratification approach with matched treatment for patients with acute back pain may lead to a very small reduction in pain levels at 3–6 months compared with usual care. This result, however, did not achieve the pre-determined clinically significant threshold.

This review also found very-low certainty evidence that risk stratification has no significant difference on disability outcomes. While the removal of Priebe et al40 during sensitivity analysis resulted in contrasting findings to the original analysis, these results would fall well short of the proposed MCID.28 Hence, overall, the use of risk stratification for adults with acute back pain may not be superior when compared with usual care, but the evidence is very uncertain. Due to the very-low certainty of evidence, future, high-quality, research may significantly change the conclusions of this review and is required to confirm the utility of risk stratification in the management of acute back pain.

There was a focus on participants at high risk of developing chronic pain in relation to their inclusion in each trial and the provision of interventions. Priebe40 was the only study to provide a different intervention for different levels of risk, including consultation with a pain specialist for those at high-risk. All other studies matched interventions for participants stratified as high-risk only (and medium-risk for Magel et al38), rather than providing different interventions to those with different levels of risk. It is possible that different conclusions could have been drawn if interventions were also developed and prescribed for those in the low and medium risk groups.

Furthermore, no study fully complied with the suggested management for each level of risk, as recommended by the authors of the STarT Back Screening Tool. The only studies which followed the recommendations for high-risk patients were Delitto et al,36 which applied a psychologically-informed intervention, and Lee et al,35 which provided an intervention based on a cognitive-behavioral approach. This lack of concordance between the care provided and suggested management for each level of risk may have impacted the results of the studies as the care provided in each study may have been less effective. Studies did not publish specific detail regarding the nature of the usual care intervention provided, therefore it is unknown whether the usual care interventions were truly comparable. It is reasonable to assume that, given the regional variation of where the studies were conducted, usual care interventions would have differed between each study.

The inclusion criteria of the individual studies may have also limited the generalizability and practicality of conclusions. All included studies took place within high-income countries, limiting generalizability to low- and middle-income countries. Lee et al35 and Magel et al,38 excluded participants with pain extending below the knees and with neurological symptoms, and the former only included participants aged 18 to 55 who had been injured at work. Magel et al38 was a secondary analysis of an RCT, and as such, the original study was not designed to evaluate the use of a risk stratification tool.

In reference to the design of this review, the exclusion of non-RCTs from the search strategy likely limited the yield of evidence. Although, to our knowledge, no other non-RCT has evaluated risk stratification specifically for acute back pain. Furthermore, the limitation of studies to only the English language may have resulted in non-English studies being missed; limiting the diversity and generalizability of findings. Additionally, there was likely a publication bias in both meta-analyses due to the small number of cohorts included.

The results of this review contrast those of a recent systematic review by Ogbeivor and Elsabbagh17 which found evidence to support the use of stratified management with matched treatment for back pain of any duration. It is worth noting that Ogbeivor and Elsabbagh17 differed from this review as it included participants with chronic pain and non-RCTs, and excluded patients stratified using tools other than the STarT Back Screening Tool. In reference to the management of acute back pain more broadly, there is high-quality evidence to support the use of exercise, heat, manual therapy, and education.42,43

While risk stratification is proposed as a method of managing acute back pain, this review found only very-low certainty evidence of improvement in pain at 3–6 months post intervention compared with usual care, and this finding did not reach clinical significance. The very-low certainty results were due to a general paucity of high-quality studies specific to only acute back pain. Future high-quality RCTs are needed to conclusively determine the effectiveness of risk stratification for patients with acute back pain, and outcomes should include the transition from acute to chronic pain.