Can we predict poor recovery from recent-onset nonspecific low back pain? A systematic review

Abstract

This systematic review of prospective cohort studies investigated the evidence for prognostic factors for poor recovery in recent-onset nonspecific low back pain (NSLBP). Medline, Cinahl, Embase, PsychINFO, and AMED databases were searched and citation tracking was performed. Fifty studies met the inclusion criteria. Bivariate and multivariable prognostic factor/outcome associations were extracted. Two reviewers independently performed data extraction and method quality assessment. Where data were available, odds ratios for bivariate associations were calculated and meta-analysis was performed on comparable prognostic factor/outcome associations. Despite the number of studies that have investigated these prognostic factors, uncertainty remains regarding which factors are associated with particular outcomes, the strength of those associations and the extent of confounding between prognostic factors. This uncertainty is the result of the disparate methods that have been used in these investigations, incomplete and contradictory findings, and an inverse relationship between study quality and the reported strength of these associations. The clinical implication is that the formation of clinically useful predictive models remains dependent on further high-quality research. The research implications are that subsequent studies can use the findings of this review to inform prognostic factor selection, and that prognostic studies would ideally be designed to enhance the capacity for findings to be pooled with those of other studies.

Introduction

Low back pain (LBP) is common, costly, diagnostically challenging, and has a variable clinical course. Rapid improvements occur in the first 3 months post-onset, but improvements are gradual thereafter (Hestbaek et al., 2003; Pengel, 2003). At 6 months following onset, 16% (range 3–40%) of patients initially off-work remain off-work, and at 12 months post-onset, 62% of all patients (range 42–75%) still have pain. Within 12 months of onset, recurrences of pain (60%, range 44–73%) and of work absence (33%, range 26–37%) are common (Hestbaek et al., 2003). Patients who do not recover early are at much greater risk of long-term poor recovery and utilize the bulk of compensable health-care resources (Torstensen et al., 1998; Van Tulder et al., 2000).

If the capacity existed to identify LBP patients at risk of poor recovery early in the clinical course of the condition, appropriate intervention might influence outcomes. Frustration experienced by patients and clinicians might be reduced by advanced warning of a potentially modest recovery, with subsequent recalibration of expectations, rehabilitation strategies, and lifestyle adjustments. Furthermore, identification of important prognostic factors for LBP patients could facilitate baseline comparability of subjects recruited into research studies. The Cochrane Back Review Group reported a need for systematic reviews to determine the most important prognostic factors for poor recovery, and referred to this as a ‘Holy Grail’ question (Bouter et al., 2003).

Several systematic reviews have examined prognostic factors for poor recovery in back pain (Rohling et al., 1995; Linton, 2000; Kuch, 2001; Teasell, 2001; Teasell and Bombardier, 2001), and for LBP specifically (Crook et al., 1998; McIntosh et al., 2000b; Truchon, 2001; Pincus et al., 2002). However, many reviews have included only workers and have measured only occupational outcomes. Some prognostic studies have used return-to-work as the sole outcome measure, and as return-to-work is influenced by many factors, it is unlikely to mirror all the dimensions of the clinical course of LBP.

Previous LBP prognostic reviews have also included cohorts with specific LBP and LBP with neurocompressive signs. Specific LBP is caused by serious pathology (cancer, spinal osteomyelitis, fracture, spinal stenosis, cauda equina, ankylosing spondylitis, visceral-referred pain). Neurocompressive signs (pins & needles, numbness and weakness) in LBP are usually caused by nerve root irritation. Specific LBP and LBP with neurocompressive signs represent approximately 20% of LBP in primary care. The remaining 80% cannot be definitively linked to a specific cause and is most accurately described as nonspecific LBP (NSLBP) (Spengler and David, 1985; Deyo et al., 1992). As there is evidence that neurocompressive signs are risk factors for chronicity (Burton et al., 1995; Hunt, 2002), it is possible that NSLBP has both a clinical course (Di Fabio, 1995) and prognostic factors that are different from neurocompressive LBP. Similarly, NSLBP may have a clinical course and prognostic factors that are different from specific LBP. No review of NSLBP has quantitatively investigated the prognostic utility of factors from the broad range of clinical assessments (physical impairment, pain, activity limitation, psychosocial function, and participation restriction) and included a comprehensive range of outcomes that are relevant to people with this condition.

Many prognostic studies have been performed in LBP but predictive models derived in these studies do not appear to be commonly used in clinical practice. Potential barriers to the clinical application of previous prognostic research may include an inadequate strength of predictive models, prognostic associations being inconsistently reproduced across studies and predictive models presented in formats not readily interpretable by clinicians.

Authorities encourage prognostic studies to be built on theoretical models and the results of previous studies (Altman and Lyman, 1998). In NSLBP there is an absence of theoretical models that traverse assessment domains. However, as there is no shortage of previous studies, a preliminary step towards optimising further prognostic studies would be to synthesize previous results. The validity of prognostic factors would be enhanced if the association with a particular outcome were reproduced across independent samples. Greater confidence in the utility of a prognostic factor might also be warranted if measures of the strength of its relationship with an outcome (such as odds ratios) could be pooled through meta-analysis.

The aim of this systematic review was to investigate the prognostic factors of poor-recovery that have been studied in recent-onset NSLBP (<12 weeks) using the outcomes of pain, activity limitation, and participation restriction. The review aimed to identify the significant prognostic factor/outcome associations that have been found, to report the predictive strength of these associations, to investigate the influence of methodological quality on these associations, and to synthesize these results or report barriers inhibiting this synthesis.