Biopsychosocial Factors for Chronicity in Individuals with Non-Specific Low Back Pain: An Umbrella Review

Low back pain (LBP) is a global and disabling problem. A considerable number of systematic reviews published over the past decade have reported a range of factors that increase the risk of chronicity due to LBP. This study summarizes up-to-date and high-level research evidence on the biopsychosocial prognostic factors of outcomes in adults with non-specific low back pain at follow-up. An umbrella review was carried out. PubMed, the Cochrane Database of Systematic Reviews, Web of Science, PsycINFO, CINAHL Plus and PEDro were searched for studies published between 1 January 2008 and 20 March 2020. Two reviewers independently screened abstracts and full texts, extracted data and assessed review quality. Fifteen systematic reviews met the eligibility criteria; all were deemed reliable according to our criteria. There were five prognostic factors with consistent evidence of association with poor acute–subacute LBP outcomes in the long term (high levels of pain intensity and disability, high emotional distress, negative recovery expectations and high physical demands at work), as well as one factor with consistent evidence of no association (low education levels). For mixed-duration LBP, there was one predictor consistently associated with poor outcomes in the long term (high pain catastrophism). We observed insufficient evidence to synthesize social factors as well as to fully assess predictors in the chronic phase of LBP. This study provides consistent evidence of the predictive value of biological and psychological factors for LBP outcomes in the long term. The identified prognostic factors should be considered for inclusion into low back pain explanatory models.


Introduction
Low back pain (LBP) is a common health condition with important implications for individuals, public health systems and economies [1]. It has been increasing worldwide since 1990 with the rise and aging of the population, with a higher prevalence among people between the ages of 40 and 80 [2,3]. In 2017, low back pain was the leading cause of years of disability, with over 570 million people affected at any one time [3], and it is likely to increase in low-income and middle-income countries in the next few decades [4]. Low back pain generates an impact on the quality of life of individuals [5,6] and on the economy, with direct healthcare costs [7] comparable to those of cardiovascular disease, cancer or mental health [8], as well as indirect costs related to the potential loss of work status [4,9].

Review Selection
We selected systematic reviews, with or without meta-analysis, summarizing longitudinal observational studies that involved adult participants (≥18 years) at any point in the course of LBP (acute, subacute or chronic) or with mixed pain (i.e., other conditions such as neck or thoracic pain), only if most of the population (≥75%) underwent NSLBP or subgroup data were available for this condition, with baseline measures of at least one biological, psychological or social factor, as well as one predicating the primary outcomes (pain intensity, functional status, work participation and recovery) and, additionally, secondary outcomes (health-related quality of life, emotional distress, satisfaction with treatment and healthcare utilization); we included only those written in English or Spanish.
We excluded reviews involving a majority of individuals with LBP caused by specific pathologies or conditions (such as surgery or pregnancy); those assessing factors as mediators, moderators or their impact on treatment; those reporting only secondary outcomes; those based on a cross-sectional design; and narrative or methodological reviews.

Data Extraction and Management
We recorded complete information about citations, populations, methods, prognostic factors and outcomes assessed. The results of the reviews were extracted separately for each duration of LBP symptoms: acute-subacute (≤3 months), chronic (>3 months) and mixed duration [20].
Likewise, results data were extracted for each primary outcome, according to the International Classification of Functioning, Disability and Health (ICF) framework [21]-pain intensity, functional status and work participation and recovery-which were considered together to synthesize the evidence of our outcomes of LBP results at follow-up. For the interpretation of the best available evidence, the secondary outcomes of health-related quality of life, emotional distress, satisfaction with treatment as well as healthcare utilization were collected and considered narratively. We categorized the results according to the follow-up time period-short-term (<3 months) and long-term (≥3 months)-along with the evidence that most improvements in pain, activity limitation and return to work occur within 3 months and thereafter recovery is lesser [20]. Moreover, since an unadjusted finding does not control for confounding factors, unlike the adjusted finding, we extracted all adjusted data, apart from the unadjusted data for a separately planned analysis, when possible [22].
When a systematic review presented data from several primary studies for the same factor, we reported the range (i.e., the lowest and highest value reported). In the event that the review described a meta-analysis, we presented the pooled estimate. Where several measurement instruments were reported for the same outcome, we selected the measure with greater evidence of validity and reliability for synthesis. Likewise, all dichotomous measures with more than one cut-off point were extracted, but the one showing the most significant association was used. In addition, the overlap of primary studies among the included reviews was recorded using citation matrices and excluded from our synthesis. The degree of overlapping studies was calculated using the Corrected Covered Area (CCA) method [23].

Methodological Quality Assessment of Included Reviews
We used the criteria developed by the SUPPORT and SURE collaborations, reported in a recent review published in The Cochrane Library [24]. It rates 14 criteria grouped under Section A-Identification, selection and critical appraisal of studies; Section B-Analysis; and Section C-Overall. Each item can be rated as follows: +, yes; ?, can't tell/partially;-, no; NA, not applicable (e.g., no studies or data). In the last item, and considering the prior assessments of the criteria, the review is categorized as having (1) only minor limitations; (2) limitations that are important enough that it would be worthwhile to search for another systematic review and to interpret the results of this review cautiously, if no better review is available; (3) limitations that are important enough to compromise the reliability of the findings of the review and to prompt the exclusion of the review.

Data Synthesis
The characteristics of the included reviews were summarized descriptively. We also conducted a descriptive quantitative analysis (summary measure with a precision estimate) for each systematic review, according to the duration of LBP symptoms (acute, subacute, chronic and mixed duration) as well as length of follow-up (short and long term).
To adequately compare findings across the reviews, we used odds ratio (and beta coefficients) statistics for synthesis. Results of a systematic review were considered consistent if ≥75% of the primary studies reporting on a factor rated the same direction of association with the outcome [25]. Thus, a factor was judged consistently associated with low back pain outcomes when it demonstrated a uniform association in the same direction by at least two reliable reviews, or at least half of them, and not contradicted by any other review [15]. The strength of association with outcomes was deemed weak (OR 1.01-1.49), moderate (OR 1.50-1.99) or strong (OR ≥ 2.0) [26], with moderate and strong strengths considered clinically relevant.
Thus, a qualitative synthesis was performed given that the main purpose of this study was to present a summary of the current body of evidence based on systematic reviews of biopsychosocial prognostic factors in patients with LBP and also considering the heterogeneity of the data collected. In this way, we have described and discussed the extent of the main differences found in the results reported by the included reviews, as well as the aspects considered as probable explanatory factors for such heterogeneity, without performing additional subgroup or sensitivity analyses.

Review Selection
A total of 2.721 citations were identified: 1.846 through electronic databases, 744 from grey literature databases and 131 from tracking citations and contact with authors. We evaluated 72 full-text publications, and 15 systematic reviews were eligible (see Figure 1). References from excluded full-text citations (n = 57) are reported in Supplementary Table S3. The conflicts of interest of the review authors are displayed in Supplementary Table S4. Disagreements were resolved by consensus among reviewers twice during the selection process, four times during data extraction and twice during quality assessment, with non-intervention of the third reviewer.
The publication date of the included reviews ranged from 2008 to 2019 and that of the included primary studies varied from 1981 to 2017 (Supplementary Table S5: Primary studies referenced in tables). Twelve reviews (80%) were published over 5 years ago (before 2015). The sample size in the reviews ranged from 219 [41] to 112.797 [29], with a mean of 17.147 (interquartile range (IQR): 3.535 to 11.330).
Regarding our outcome of LBP results, all primary outcomes were widely assessed across the 15 included reviews: work participation (60%, 9 reviews), functional status (60%, 9 reviews), pain intensity (60%, 9 reviews) and aspects of recovery (53%, 8 reviews). For secondary outcomes, only satisfaction with treatment and healthcare utilization results were reported by two primary studies, using p-values in two reviews [37,39], with insufficient evidence for interpretation. Only 5% of the primary studies included in the systematic reviews reported results within 3 months of follow-up (short term).
The main reasons for the review authors not pooling the results were the heterogeneity of the population, measures of prognostic factors, outcomes assessed and outcome measures, as well as the variety of statistical analyses. The most common estimates used to report the results across the reviews were odds ratios (OR), but beta coefficients (β), risk ratios (RR), prevalence ratios (RP), hazard ratios (HR), likelihood ratios (LR+/LR-) and p-values were also reported. Data from RR/RP, HR, LR+/LR-and p-values are provided in Supplementary Table S6.

Methodological Quality Assessment of Included Reviews
The results of our appraisal of the methodological quality (reliability) of the included reviews are shown in Supplementary Table S7. We judged all 15 included reviews to have only minor limitations. In general, there were few failures with regard to the selection criteria and critical appraisal of the risk of bias of the primary studies from the systematic reviews, with thirteen reviews partially meeting the comprehensive search strategy criterion. Likewise, there were few flaws regarding the analysis of the results, with three reviews showing limitations in the reporting of the characteristics and results and one review explaining the differences in the results. Disagreements were resolved by consensus among reviewers twice during the selection process, four times during data extraction and twice during quality assessment, with non-intervention of the third reviewer.       NSLBP = non-specific low back pain; NR = not reported; RQ = review search question; SS = statistically significant; NS = not statistically significant; CI = confidence interval; S/T = short term; L/T = long term; TrA: transversus abdominis; LM: lumbar multifidus.

Synthesis of Results
Overall, forty-nine factors were reported across the included systematic reviews. The degree of overlap of the primary studies through the reviews was slight (CCA = 2.6%).

Acute-Subacute Phase of LBP
For most factors collected, there was insufficient evidence to synthesize the unadjusted data, so our summary of results was mainly based on adjusted results. Besides this, there were three factors (gender, previous history of LBP and pain radiating to the leg) for which there was also insufficient evidence to perform synthesis from adjusted data. In order to include as much evidence as possible [22], we combined both types of results (adjusted and unadjusted) to analyze the consistency of these variables.
Thus, there were 10 prognostic factors of outcomes at long-term follow-up provided by two or more systematic reviews with sufficiently similar data for comparability (OR/Beta), derived from seven systematic reviews (Tables 2 and 3) [27,29,30,32,33,35,38]. Of these, five prognostic factors showed consistent evidence supporting their ability to predict poor long-term outcomes: high levels of pain intensity and disability, high emotional distress, negative recovery expectations and high work physical demands (Table 4). Another factor showed consistent evidence of no association with poor outcomes: low education levels. Each of these variables showed strengths of association ranging from weak to strong and outcomes reflecting clinical relevance (OR ≥ 1.50). Moreover, four factors demonstrated no consistent evidence supporting their predictive ability for long-term outcomes: high fear avoidance beliefs (from adjusted data) and female gender, the presence of previous history of LBP and pain radiating to the leg (from adjusted and unadjusted data) (Tables 2  and 3). These variables did not reflect relevant disagreements of inverse association. On the other hand, there were 35 prognostic factors reported by a single systematic review, with insufficient evidence for synthesis (Supplementary Table S8).

Chronic Phase of LBP
There were four variables reported by a single systematic review and, therefore, with insufficient evidence for synthesis [35,36,40,41]: physical activity, abdominal muscle function, fear avoidance beliefs and pain catastrophism. The evidence in all of them ranged from non-association to association with the results (Supplementary Table S9). There was only one finding indicating that high fear avoidance beliefs predicted better low back pain outcomes. However, the sample size of this study was small and the follow-up short, so this may be a potentially biased finding.

Mixed-Duration LBP
Pain catastrophism was reported by two reviews based on individuals with acute to chronic LBP [36,37], reflecting that high catastrophic thinking showed a consistent association with poor long-term outcomes and clinically relevant strengths of association (from adjusted data) (Table 4). Moreover, there were three factors in acute to chronic LBP [31,35,40] and one factor in the subacute-chronic population [34] reported by a single systematic review: physical activity, fear avoidance beliefs, work social support and recovery expectations, respectively. Once again, the evidence ranged from association to no significant association with outcomes in each one of them (Supplementary Table S10).
On the other hand, only four factors were analyzed in the acute, subacute and chronic low back pain phases. Recovery expectations were systematically associated with outcomes regardless of the duration of symptoms, and pain catastrophism showed a trend towards association in all phases, although not always significantly. Physical activity showed a tendency of non-association in the different phases, and fear avoidance beliefs were more significant in the subacute phase of LBP.
The considerable clinical and methodological heterogeneity of the collected data (prognostic factors, outcomes and their measurements, as well as the wide range of time in the long-term follow-up) precluded the use of a meta-analysis.      Supplementary Table S5; OR = odds ratio; NR = not reported; NA = not assessable; ss = significant result; ns = non-significant result. Outcome: P = pain; FS = functional status; WP = work participation; R (a, b, c, d): recovery a = recovery of pain or disability, b = self-reported recovery, c = slightly better" or "worse" score on two or more follow-up measurements, d = recovery and/or return to work. * Sample of individuals in acute phase of low back pain. ** Sample of individuals in subacute phase of low back pain. *** Meta-analysis combining adjusted and adjusted data. Bold results are statistically significant.

Discussion
This umbrella review provides a summary of up-to-date and high-level research evidence about biopsychosocial predictors in individuals with NSLBP. We included 15 systematic reviews, showing primary research spanning the last three decades.
A variety of biopsychosocial prognostic factors have been investigated but, in accordance with the evidence derived from the present umbrella review, only high levels of pain intensity and disability, high emotional distress, negative recovery expectations, high pain catastrophism and high work physical demands are predictors of poor low back pain outcomes at long term, and low levels of education have no prognostic ability.

Acute-Subacute LBP
In the present umbrella review, the factors found to be associated with poor outcomes in this phase of LBP are largely in line with the literature on LBP [15,[44][45][46][47][48] and MSK [49][50][51] prognosis. In spite of this, we consider that the results suggesting that high baseline pain intensity and disability levels predict LBP outcomes should be understood from the perspective of their interactions with the factors that we discuss below. We found that individuals with high levels of emotional distress are at a greater risk of developing chronic pain and disability, with depression being the predictor with the greatest strength of association. However, its predictive capacity for the maintenance of chronic low back pain, beyond its association derived from cross-sectional studies, has been less reported in longitudinal studies, as this umbrella review shows. Nevertheless, a recent review with qualitative data on chronic LBP showed that depression had moderate evidence of no association with work-related outcomes at follow-up [52]. Moreover, recovery expectations were the most consistently reported predictor in the current umbrella review, regardless of the different outcome domains considered, as well as the phase of low back pain analyzed. Similar results have been reported in individuals with conditions other than back pain, including chronic shoulder pain [53] and major orthopedic trauma [54]. In addition, we mainly found strong association strengths with poor outcomes for high work physical demands, indicating the clinical relevance of this factor in individuals with acute-subacute LBP, in line with the previous overview of LBP prognostic factors [15]. However, two recent reviews in populations with MSK pain found insufficient evidence for physical workload [49,51] that may suggest the greater relevance of these aspects for the low back region specifically.

Mixed Duration of LBP
Despite our results reflecting consistent evidence that high pain catastrophism predicts a delay in the functional recovery of individuals with acute to chronic LBP, as well as a trend of association with poor outcomes in the other phases of LBP, the role of catastrophic thinking remains controversial. A systematic review and meta-analysis of mediation studies suggested that "catastrophism may not explain the development of disability from back and neck pain" [55]. Moreover, it has been recently reported that pain-related acceptance is a significant mediator both between pain and catastrophism and between catastrophism and fear avoidance beliefs in chronic pain patients [56]. Thus, more studies are needed to understand the cognitive processes in the experience of pain.

Factors with Consistent Evidence of No Association with Poor Outcomes at Long Term in
Acute-Subacute LBP We found that a lower level of education was not associated with worse work-related outcomes, being in line with the evidence provided by previous reviews in LBP [15,48,52] and musculoskeletal populations [49].

Factors with Inconsistent Evidence of Association with Poor Outcomes at Long Term in Acute-Subacute LBP
The inconsistent evidence found for the female gender was mainly due to the findings reported by Agnello et al., but whose significant heterogeneity was explained by the compensation status of the individuals to participate in the study [30]. Considering this, our findings are consistent with the non-association evidence reported by other authors in LBP and MSK pain [15,49,51]. On the other hand, sciatica or nerve root exam results showed consistent evidence of association with poor acute-subacute LBP outcomes in a previous overview [15]. Our findings of inconsistent evidence for pain radiating to the leg could be related to the fact that the included reviews did not provide an explicit definition and their measurements ranged from LBP assessment with or without radiating pain to the assessment of neurocompressive radiculopathy. Moreover, in both the current umbrella review and the prior overview of prognostic factors in LBP [15], having previous episodes of low back pain showed inconsistent evidence of association with acute-subacute low back pain outcomes. The lack of consensus in the definition of recurrence versus new episodes of LBP [57] could explain in part the lack of consistency in these findings. Finally, the predictive role of fear avoidance beliefs (FABs) in the development and perpetuation of chronic pain has been systematically reviewed in samples of LBP [15] and musculoskeletal pain patients [44,58], with some conflicting results between them as well as with the present umbrella review. The concepts of fear and avoidance encompass a series of complex processes that interact over time, and this may suggest that they are linked. However, pain-related fear and avoidance behaviors are context-dependent and do not always cooccur [11]. Thus, an individual can both prioritize the goal of avoiding pain for protection, even without reporting fear [59], and can prioritize other valued life goals and confront the threat whilst self-reporting fear [11]. This confusing conception of fear related to pain and avoidance behaviors, evidenced in turn through the measurement instruments available so far [60], may partially explain the conflicting evidence found in this umbrella review, reflecting the complexity of these mechanisms.

Other Factors with Insufficient Evidence of Association with LBP Outcomes at Long Term
In the current umbrella review, low work social support [31] and low social activity [27] were reported by one systematic review, showing predictive ability for poor outcomes in individuals with mixed-duration and acute-subacute LBP, respectively. A recent systematic review among individuals with chronic pain found that the most frequent aspect in explaining the effect of social support on the experience of pain was the stress-buffering effect [61]. More studies analyzing the mechanisms of interaction between social factors and disabling LBP are needed. In addition, older age is considered a common predictor of poor outcomes in LBP, musculoskeletal pain and sciatica [15,51,62]. We believe that age may influence the natural course of low back pain and more studies are needed to determine its predictive value in these individuals.

Strengths and Weaknesses
We developed and registered a specific overview protocol in PROSPERO, minimizing reporting bias and giving transparency to the review process. Our search strategy was implemented in a sufficiently inclusive manner through relevant and grey literature databases, along with additional strategies such as manual searches and contact with authors (accounting for 14% of the reviews included), reflecting the evidence from original studies over the last 35 years and including a large number of participants (N = 257,208).
The weaknesses of the present overview depend not only on the risk of bias and selective reporting of results by the primary studies, as reflected the publication biases shown in the findings derived from meta-analyses, but also on the quality of the included reviews, all of them being assessed to have minor limitations. Additionally, there was a modification from the initial protocol recorded in PROSPERO. For our outcome of LBP results at follow-up, we planned to synthesize the evidence for each primary outcome separately, but, due to insufficient evidence, we considered pain intensity, functional status, work participation and recovery outcomes together. Furthermore, at the level of this overview, the English and Spanish languages were considered as inclusion criteria, and therefore some reviews of interest may have been excluded. Moreover, the heterogeneity derived from the variability in adjustment models for confounders must be recognized. Our synthesis is also limited by the fact that we only included quantitative research studies; for this reason, several systematic reviews with qualitative data have been considered in our discussion.

Implications for Clinicians and Policymakers
This umbrella review presents a synthesis of prognosis evidence on individuals with acute and subacute LBP in North America, Europe and Oceania in clinical and occupational settings. An enhanced understanding of the role of the psychosocial factors provides the opportunity for prevention, identifying patients at risk of chronicity and targeting treatments for modifiable factors [63][64][65]. Treatments in low back pain may consider the factors consistently reported in this umbrella review. Policymakers should include multidimensional interventions through public health systems [66].

Future Research
The factors presented in the present umbrella review, with consistent evidence of a prognostic association with LBP outcomes derived from adjusted data, can be taken into consideration for the development of low back pain causal explanatory models and for intervention trials in these patients. In view of the evidence collected, further research in the later phase of LBP and regarding social and socio-occupational factors is required. Future reviews that include a meta-analysis could gain a better estimate of prognostic effect sizes, assess and account for heterogeneity in the effects of prognostic factors and perform additional subgroup and sensitivity analyses.
Overall, we still need a better understanding of the complex dynamic relationships between biopsychosocial factors.

Conclusions
The current umbrella review has identified consistent findings of up-to-date and highlevel research evidence that support the ability of several biopsychological factors to predict LBP outcomes in the long term. Such factors are levels of pain intensity and disability, emotional distress, recovery expectations, pain catastrophism and physical demands at work. These variables deserve attention for inclusion in the development of low back pain explanatory models. More research on social and socio-occupational factors, as well as predictors, in the chronic phase of LBP is required in order to add potential prognostic information to this condition. Our findings implicate a multidimensional approach in dealing with these individuals.  Table S5: Primary studies referenced in tables; Table S6: Other statistics (RR, RP, HR, LR+LR-and p values); Table S7: Reliability of included reviews; Table S8: Prognostic  factors reported by only one review; Table S9: Prognostic factors in chronic LBP; Table S10