What have we learned from ten years of trajectory research in low back pain?
Aron S. Downie 5 , 6 and
Kate M. Dunn 7
BMC Musculoskeletal DisordersBMC series – open, inclusive and trusted2016
Non-specific low back pain (LBP) is often categorised as acute, subacute or chronic by focusing on the duration of the current episode. However, more than twenty years ago this concept was challenged by a recognition that LBP is often an episodic condition. This episodic nature also means that the course of LBP is not well described by an overall population mean. Therefore, studies have investigated if specific LBP trajectories could be identified which better reflect individuals’ course patterns. Following a pioneering study into LBP trajectories published by Dunn et al. in 2006, a number of subsequent studies have also identified LBP trajectories and it is timely to provide an overview of their findings and discuss how insights into these trajectories may be helpful for improving our understanding of LBP and its clinical management.
LBP trajectories in adults have been identified by data driven approaches in ten cohorts, and these have consistently demonstrated that different trajectory patterns exist. Despite some differences between studies, common trajectories have been identified across settings and countries, which have associations with a number of patient characteristics from different health domains. One study has demonstrated that in many people such trajectories are stable over several years. LBP trajectories seem to be recognisable by patients, and appealing to clinicians, and we discuss their potential usefulness as prognostic factors, effect moderators, and as a tool to support communication with patients.
Investigations of trajectories underpin the notion that differentiation between acute and chronic LBP is overly simplistic, and we believe it is time to shift from this paradigm to one that focuses on trajectories over time. We suggest that trajectory patterns may represent practical phenotypes of LBP that could improve the clinical dialogue with patients, and might have a potential for supporting clinical decision making, but their usefulness is still underexplored.
Decisions about health care are traditionally based on a medical diagnosis. However, the most important focus of health care is patient outcomes and, as recently argued by Croft and colleagues, these outcomes are not only determined by disease diagnosis [ 1 ]. Sometimes diagnosis actually tells very little about prognosis. Croft and colleagues argue that “prognosis can now provide the framework in which clinicians and researchers organise evidence and information to support decisions about management”, and illustrate this proposition with numerous examples of prognostic factors being fundamental for clinical decisions [ 1 ].
Low back pain (LBP) is a health condition in which diagnostic information usually does not tell much about probable future outcomes, as in only a minority of cases can a specific pathoanatomic diagnosis be reached [ 2 ]. The majority of LBP is categorised as non-specific LBP and therefore may be better understood and managed within a prognostic framework [ 1 ].
Non-specific LBP is often categorised as acute, subacute or chronic focusing on the duration of the present episode [ 3 ]. However, more than twenty years ago it was recognised that LBP is often an episodic condition and people who have experienced LBP are likely to also have future episodes [ 4 , 5 ]. This challenged the concept of acute versus chronic LBP which implies that LBP presents either as unrelated acute episodes or as chronic continuous pain, and an additional limitation of that concept is that it does not differentiate between a recent onset episode experienced for the first time and a recent flare-up of recurrent LBP. Similarly, this categorisation of chronic LBP includes both people with persistent severe pain and people reporting mild symptoms for more than three months.
That LBP often presents as recurrent episodes also implies that the population-averaged course of LBP does not adequately reflect the course experienced by individuals. The averaged course of LBP with early improvement, followed by very little change after 6 to 12 weeks [ 6 , 7 ], has been translated into a perception of LBP as a condition that is largely unchanged after that time period. However, recognising that LBP comes and goes suggests that it is not the same individuals reporting pain at all time-points.
Despite this, it was not until ten years ago that patterns underlying the averaged course of LBP were investigated. A longitudinal observational study with monthly follow-up measurements showed that distinct trajectories of LBP (patterns of changes in pain over time) could be identified, and indicated that the prognosis of LBP cannot be adequately described in terms of simply recovery or chronicity [ 8 ] and also that a population-averaged course of LBP does not adequately reflect the underlying patterns of LBP. Since that pioneering study, trajectory patterns have been identified in a number of cohorts from different settings and countries and using different statistical methods. These have all confirmed that characteristic LBP prognostic groups exist with trajectory patterns that are distinctly different from the population-averaged course. A recent overview of LBP trajectory studies concluded that most people who experience LBP will have trajectories of either persistent or episodic pain rather than one well-defined episode, and suggested that single time-point outcomes are not optimal measures of LBP [ 9 ].
In this paper, we provide a summary of the current knowledge on LBP trajectory patterns in adults and describe the main similarities and differences of previous findings. Subsequently, we consider whether LBP trajectory patterns may be useful as a way to define LBP prognostic ‘phenotypes’. Lastly, we discuss how such trajectories may become clinically useful and suggest some areas for future research.
Which LBP trajectories have been identified?
To our knowledge, LBP trajectories in adults have so far been identified by data driven approaches in ten cohorts [ 8 , 10 – 18 ]. In these studies, participants with a main complaint of LBP were followed from three to twelve months with data collection at four to 52 time-points. Outcome measures were LBP intensity, LBP frequency (number of LBP days per week) and activity limitation. Trajectory patterns were identified using either Hierarchical Cluster Analysis, Latent Class Analysis, or Latent Class Growth Analysis (Table 1 ).
Overview of ten studies in which LBP trajectories have been identified by data-driven approaches
Timing and duration of follow-up
n = 342 (2001–03) [ 8 ]
n = 155 (2009–10) [ 10 ]
Hierarchical Cluster informed by spline regression (intercept, slopes, knot)
Typical [improve markedly during 4 weeks] 41 %
Stable [mild] 24 %
Primary care + Secondary care [ 16 ]
General practice + outpatient clinic
informed by linear regression (deviations from line)
Non-fluctuating 87 %
Workers on sick leave [ 13 ]
n = 678
Hierarchical Cluster informed by linear regression (slope)
Continuous high 42 %
Primary care + emergency care [ 17 ]
Age >65 years
General practice (